symanter standardizing storage management 03 2006.en-us

8/8/2019 Symanter Standardizing Storage Management 03 2006.en-us

1/167

Sean Derrington

Josh Kahn

Paul Massiglia

Standardizing StorageManagement

Using the Veritas Storage Foundation to increase

the effectiveness of storage management across

the enterprise


2/167

Copyright 2006 Symantec Corporation.

All rights reserved.

Symantec and the Symantec Logo are trademarks or registeredSymantec Corporation or its affiliates in theU.S. andother countmay be trademarks of their respective owners.

No part of the contents of this book may be reproduced or transm

or by any means without the written permission of the publishTHE DOCUMENTATION IS PROVIDED "AS IS" AND ALL EXPRECONDITIONS, REPRESENTATIONS AND WARRANTIES, INCLUIMPLIED WARRANTY OF MERCHANTABILITY, FITNESS FOR APURPOSE OR NON-INFRINGEMENT, ARE DISCLAIMED, EXCEPTTHAT SUCH DISCLAIMERS ARE HELD TO BE LEGALLY INVALCORPORATION SHALL NOT BE LIABLE FOR INCIDENTAL OR C

DAMAGES IN CONNECTION WITH THE FURNISHING OR USEDOCUMENTATION.THEINFORMATIONCONTAINEDIN THISDIS SUBJECT TO CHANGE WITHOUT NOTICE.

Symantec Corporation 20330 Stevens Creek Blvd. Cupertino, CA

http://www.symantec.com

ISBN 0-321-447794


3/167

AcknowledgmentsOur goal in writing this book has been to articulate the value propostorage management standardization, and to show how the Veritas Foundation by Symantec can make standardized management of ona reality in large and heterogeneous data centers. We were aided in Sean Derrington, of Symantecs Data Center Management Group, wh

the basic ideas of standardization for us.We are particularly grateful to Manish Parekh and Aglaia Kong of thFoundation Management Server group for access to test systems fordeveloping Chapter 6, as well as to Distinguished Engineers Craig HRon Karr for their painstaking technicalreviews of theVxFSfile systevolume manager material respectively. Also to DCMG Technology SJohn Colgrove, who continually remindedus that this is the most imof the series.

As any author knows, creating the content of a book is only part of tWe thank Sarah Connell-Smith of the Symantec Infodev organizatiothird great Yellow Book editing job in as many months. LMA Associaquick and accurate turnaround of the graphics. Symantecs productiparticular, Jim Hudgens, Fran Heddings, and Rossana Monzon did qaccurate conversions from raw manuscript to camera-ready images,cover art in next to no time, all while they were trying to be accomm

undisciplined and tardy authors.

Finally, our special thanks to our project manager Kristine Mitchellpersistence and good humor as we all learned together how Symantbook from raw manuscript to finished product are responsible in larfor Standardizing Storage Managementfinding its way into your han

If despite the extraordinary efforts of this team, errors and omissioresponsibility for them lies solely with the authors.

Sean DerringtonJosh KahnPaul Massiglia

Mountain View, CAApril 2006


4/167


5/167

Chapter 1 The what and why of standardized storagemanagement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 2 The Storage Foundation virtualization platform

Chapter 3 A closer look at the Storage Foundation core: tVxVM Volume Manager . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 4 A closer look at the Storage Foundation core: tVxFS File System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 5 A closer look at the Storage Foundation: DynamMulti-pathing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 6 A closer look at the Storage Foundation: the StorFoundation Management Server . . . . . . . . . . . . . . . . . . . . . . .

Chapter 7 Using the Storage Foundation . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Chapter 8 Making standardization happen . . . . . . . . . . . . . . . . . . . . . . . . .

Contents


6/167

Contents4


7/167

The what and why ofstandardized storagemanagement

This chapter includes the following topics:

Why manage storage?

What is storage management?

The Veritas Storage Foundation by Symantec

Why manage storage?Storage is free!

This obviously overblown argument is heard frequently in manpersonal computer users to enterprise storage system salespeo

designers bent on storing every bit of data that might, just posssome day. The usual justification is to point to the dramatic anddecrease in magnetic disk prices, and in the case of the data cennetwork interconnect technologies that promote large scale flexand the availability of multiple tiers of cost-differentiated sto

Chapter


8/167

Storage consumption is growing, and the growth shows no signsSo much business value can be derived from storing and analyzinso many topics that storage consumption is increasing faster thais decreasing.

Storage management is personnel-intensive, and personnel is thinformation technology cost that increases rather than decreasiStorage must be managed, and larger amounts of storage require

management. Hardware must be inventoried, repaired when it bperiodically replaced withnewer models. Rawdisk capacity must band provisioned to application servers (which are also growing inThe right data must be placed on the right type of storage demust be protected against a variety of threats to loss or destructmanagement is personnel-intensive, and personnel is the one intechnology cost that can be relied upon to increase rather than dtime.

Online storage growth

The analyst group Forrester Research had this to say in its US infortechnology spending forecast for the second quarter of 2005:

The strongest budget growth in 2005 and 2006 will be computer equ

USspending in this areawill rise 12% in2005and 13% in2006. The

result of investments in personal computers, storage hardware, andbin2003,2004,andsofar in2005,ascompaniesupgraded their computi

after the fallowperiod of the tech depression from2001 to 2002.(Se2005, US IT Spending In Q2 2005: Still Tracking Forecast, by AndrQuoted by permission.)

Forresters European outlook was similar:

When lookingat hardware investments, only storage and server har

generateincreaseddemandformorethan30%offirms.Comparedwith

for IT services like consulting or application management outsourcin

moderate.( December 20, 2005 Europes IT Spending Outlook For Manuel ngel Mndez. Quoted by permission.)

These, and similar forecasts, suggest strongly that the amount of st

The what and why of standardized storage managementWhy manage storage?

8


9/167

is to where it is needed with the click of a mouse rather thanreconfiguration allows enterprises to plan for requirementsunpredictable at best.

Using cost-differentiated tiers of storage hardware to reducecost. Both the technology in a storage system and the configcapacity is delivered to users can affect the cost of effective cstorage, for example, costs at least twice as much per gigaby

non-redundant storage. Similarly, capacity delivered via thedrives that are appearing in so-called nearline storage sysfraction of the same capacity delivered using high-performa

Both of these techniques are undeniably effective, but both impmust be managed. To shift storage capacity from servers whereothers where it is needed, administrators must have an accurate,picture of how all storage in the data center is deployed. With in

arrays capable of presenting upwards of a thousand logical unitdata centers moving from a few enterprise-class systems to largsingle-application servers, tracking how storage is configured aused by applications is a full-time job, requiring multiple admindata centers.

Similarly, deploying multiple tiers of storage is effective at reduFor example, if 80% of a data centers onlinedata canjustifiablyconfigured as RAID5 rather than mirrored, overall storage cost by 30% with no hardware configuration change. If low-cost sedrives can be phased in to store less-critical data, average storareduced still further. But in achieving these cost savings, adminensure that critical data is placed on redundant storage devicesfrequently-accessed data is placed on high-performing devicesfiles in hundreds of file systems serving dozens of applications, aof individual files constantly changing, using storage tiers effec

non-trivial management challenge. These, and other tasks relamanagement of large amounts of storage and large numbers ofthe second major storage cost factorthe cost of administeringdata on it.

The human cost of managing storage and data

The what and why of standardized stWh


10/167

hardware replacement and the consequent migration of data to newis necessary.

The data on storage devices must also be managed. It must be backedto protect against complete loss; periodic disk-based snapshotsof critmust be taken to provide fast recovery from corruption if it is ever rmust be relocated when their business value changes so that the coa file online remains commensurate with its business value, and so

As data centers grow and become more complex internally, the leverequired to manage storage and data increases as well. In fact, data carchitecture and management have both emerged as separately ideninformation technology career specialties. The industry analysis weSearchStorage.com had this to say in its 2005 annual salary survey:

Although ITmanagement jobs continue to reign as the highest-payin

across the technology career spectrum, storage-related positions are

make inroads aswell, reflecting the continued importance of data maacross corporate America. Data has continued its explosive growth a

enterprises, and companies have respondedby hiring in sectors that

tomanaging and organizing corporate data, particularly in the stora

According to theTechTargetannualSalarySurvey,whichdrawsonsa

from across the TechTarget family ofWeb sites, two new entrants into

list of titlesare storage-related. Storagearchitect, which has leapt fro

score the number three spot, behind only CIOs and Internet architectaveragesalaryof $83,368, andaverage totalcompensationof$88,763

of storage manager has alsoappeared on the list for the first time. Co

number8, thejobhasanaveragesalaryof$74,400,withaverage totalc

coming in at $76,300.

The rising importance of storage-related jobs makes sense, as compa

to realize the importance ofmaintaining a solid and reliable infrastr

are investing accordingly.(http://searchstorage.techtarget.com/originalContent/0,289142,sid5_gci100

This report suggests both that staff positions specifically responsibstorage and management of data are appearing in IT organizations,enterprises value the positions highly, and compensate them corres

The what and why of standardized storage managementWhy manage storage?

10
http://searchstorage.techtarget.com/originalContent/0,289142,sid5_gci1002817,00.htmlhttp://searchstorage.techtarget.com/originalContent/0,289142,sid5_gci1002817,00.html


11/167

What is storage management?It is fashionable to measure storage management cost in terms terabytes that can be managed by a single administrator. Whilejustification for this approach, it tends to understate the full comanagement by relegating certain important tasks, such as bacsnapshot management, file placement and relocation, and so foapplication or system management domain rather than the stordomain. A more accurateway of assessing the human cost of stowould be to classify the administrative tasks it entails, regardlesthem, and explore possibilities for simplifying them.

The administrative tasks that fall under the heading of storagegenerally deal with a range of objects:

Storage hardware devices (disk drives and disk arrays)

Storage network components (switches, routers, and host bu

Virtual storage, whether host-, network-, or disk array-base

File systems and the files they contain

One can argue that databases, as repositories of digital informatype of storage management object. Although database adminiwell-developed IT specialty in its own right, it is, in fact, postul

assumptionthat files,virtual storage devices,and disk drives andused to hold database data.

The impact of the horizontal information technology indu

Because of the way in which the enterprise information technocommunity has evolved in recent years, each layer of the storaghardware, network components, and software, is usually delive

vendors, each of whom strives to be best-in-class for the layerproducts. Worse yet, some enterprises select and deploy storagbasis rather than for the enterprise as a whole. In this environmeffectively act as their own system integrators, at least to someAdditionally, each element in the storage-application chain is tb t l li d b it d b thi d t If d t

The what and why of standardized stWhat is sto


12/167

A service and sparing contract with the vendor, including an enfservice level agreement

Training and on-going expertise in user-level management operas LUN configuration

Similarly, for each system platform in the data center, the enterpris

A service and sparing contract with the vendor, including an enf

service level agreement Training and on-going expertise in general system administratio

Storage-specific training and on-going expertise in platform venmanagement tools (for example, logical disk managers, native fiand so forth)

Finally, and perhaps most expensively, for each specialized storage software package in use in the data center, such as volume manager

managers,replicators, storage networkpath managers, and so forth, tneeds:

A service contract with the vendor, including an enforceable servagreement

Training and on-going expertise in the use of the package on eachardware on which it runs or with which it interfaces

Even from this brief list it is clear that enterprise storage managema significant number of relationships with hardware and software vwhich consume administrative time and effort. Maintaining up-to-din a broad range of technologies consumes even more administrativeffort. Intuitively, it seems that reducing the number of vendors predata center, or looked at another way, standardizing the componentdeliver common IT services like storage, would tend to reduce the hstorage management, thereby increasing the number of terabytes th

administrator could manage effectively.

Approaches to reducing storage management complexityThere are two basic approaches to reducing the complexity caused bof vendors present in a data center:

The what and why of standardized storage managementWhat is storage management?

12


13/167

Reducing the number of vendors present in the data center seefirst glance. The fewer vendors that are present, the more equipeach vendor supplies, and the more important the enterprise bvendor as a customer. But reducing the number of vendors has ausually called vendor lock-in. The more dependent on a venddatacenter becomes, the more difficult and disruptive it is to dispproducts and services in favor of lower cost, more advanced tecadvantageous relationships with other vendors. Arguably, this part of users is what has led to the evolution of the informationindustry into horizontal tiers of best in class vendors.

In the long term, a more effective approach to minimizing storacomplexity and cost is to seek outhardware and software compona wide range of management capabilities for the majority of thestorage platforms in use in a data center. The recent trend towavirtualization is a manifestation of this. The server virtualizati

create a number of equivalent system environments, often runnLinux, that can be deployed interchangeably as needed in deparlimited-scale applications.

But not all applications run on all platforms, and the most crititend to require the power and robustness of enterprise-class sepresence of multiple computing platforms in enterprise data cefor the foreseeable future.

For storage itself, standardization would seem to be simpler. Thof storage is to provide raw capacity for cost. And indeed, manystandardize on a primary storage vendor with a secondary vendagainst lagging technology or disadvantageous pricing. But recsystem vendors have introduced disk arrays that utilize desktoto reduce the cost of keeping less-critical data online. This has among enterprises to adopt multi-tier storage strategies. To r

storage cost, enterprises accept the burden of more suppliers orlines, and the administrative complexity of getting the right datype of storage device.

But the greatest source of storage management complexity liesstorage network management software. Storage and network sut l f i th t th l I t ti f th

The what and why of standardized stWhat is sto


14/167

host-based or disk array-based replication), and which vendors softwill be used to perform them.

Figure 1-1 The storage management complexity issue

ServiceGuard PolyServe

Sun Cluster MSCS TrueCluster GeoSpan

ClusterFrame HACMP

Clustering

Snapshots

Volume Management

Replication

File Sys

SAN Copy SnapShot

SnapView Snap

Instantlmag FlashCopy ShadowCopy TimeFinder

MirrorDisk -UX

SVC LVM

ASM DLM

LDM SVM

EVM SDS

TrueCopy MirrorView

DoubleTake RepliStor

PPRC Shadowlmag SRDF SNDR

Data Replication Manager

JFS2

GPFS

SAM FS

Ext3

SAN-FS

OCFS

MPxIO

HDLM

SecurePath

Multipa

Thus, the storage management situation for most enterprise data cesuggested graphically in Figure 1-1 may be summarized as follows:

Multiple computing and storage system platforms are present inand will continue to be present for the foreseeable future for a vareasons. Important applications are platform-specific; mergers corganizations with different standards and philosophies, new pltechnology phases into data centers gradually, and so forth. Diffi t biliti d d t il bilit k t l h

The what and why of standardized storage managementWhat is storage management?

14


15/167

Each tool incurs incremental management cost. In the shortadministrators must be trained to use it. In the longer term,must maintain an administrative presence skilled in operatias long as it is in use.

The ideal storage management toolsetIntuitively, it seems that an effective way to attack storage man

would be to seek out the most comprehensive set of storage mathat provide:

Complete functionality, both for basic functions like virtualsystem management, as well as advanced capabilities like snreplication,storagenetwork path management, multi-tier storand so forth

A range of system and storage platform support, so that admlearn techniques once and apply the knowledge many times

Integration up and down the application-storage stack, so thtwo or more layers, such as file system expansion or shrinkaaccomplished in one administrative operation rather than s

Integration with adjacent software, such as backup managermanagement systems, and low-level disk array, host bus ada

network management tools

The Veritas Storage Foundation by SymanteThe properties of the ideal storage management toolset enumepreceding section are, in fact, the properties of Symantecs VeriFoundation. The Storage Foundation, described in more detail i

consists of a core host-based volume manager (VxVM), and file well as advanced tools for central management of storage hardwassets (CommandCentral) and the objects created by the Storage(Storage Foundation Management Server). In addition to comprvirtualization and file management functionality, the StorageFoadvancedcapabilities such as host-based remotedata replication

The what and why of standardized stThe Veritas Storage Found


16/167

thelow-level management tools supplied by allmajor storageand stohardware vendors, thereby providing complete visibility of storage network resources all the way from the application to the disk driveLUN. Together, the components of the Storage Foundation comprisemost effective means of reducing storage management complexity aavailable to enterprises today.

The Storage Foundation value propositionStandardizing on the Veritas Storage Foundation for all UNIX platfocenter attacks the problem of storage management complexity and

One set of tools for managing storage systems and network comall UNIX platforms means that administrators train for and maiadministrative skills once rather than once per computing platfosystem type.

One set of tools means that administrators are more versatile becan manage storage for any platform in the data center.

One set of tools for managing storage means greater administratoand consequently more efficient execution with fewer errors.

Oneset of tools means standardized storagemanagement proceduhardware components are present in the data center.

One set of tools means lower acquisition cost for storage managem

One set of tools means that its easier and less disruptive to taketechnology and pricing trends by introducing new hardware comthe data center.

Oneset of tools optimizes storage hardwareinvestments, for examolder disk arrays as replication targets, or by mirroring betweendisk arrays.

One set of tools means not having to maintain separate softwarevendor relationships for every storage administration function icenter.

In addition to reducing the cost and complexity of basic storage admfunctions the Storage Foundation opens up new possibilities for util

The what and why of standardized storage managementThe Veritas Storage Foundation by Symantec

16


17/167

StorageFoundation heterogeneousmirroringand storagenetwfacilities simplify the migration of data between different ty

Storage Foundation Portable Data Containers make it possibgenerated on one type of UNIX platform directly on a differor Linux platform without network copying or cross-mount

Storage Foundation Intelligent Storage Provisioning (ISP) exadministrators scope by automating most routine storage p

operations, while preserving enterprise intent for data availperformance.

Storage Foundation Management Servercentralizes the manstorage, network connections, and file systems for the entirea single console,an administrator can oversee allstorage objeto individual ones to manage reconfigurations, expansions, so forth, as necessary.

Storage Foundation Volume Server provides data center-widhost-based virtual storage, taking a big step toward data shaof different types.

The Storage Foundation for Databases family integrates basFoundation capabilities withmajor databaseplatforms includRAC, DB2, and Sybase, providing advanced storage services,and automatic file relocation, directly to the database admin

This book presents the case for Symantecs Veritas Storage Fou3 and 4 discuss the VxVM and VxFS technologies that make up Foundation core. Chapters 5 and 6 discuss Dynamic Multi-pathStorage Foundation Management Server (SFMS) respectively. Ccommon use cases that demonstrate the value of the Storage Foeveryday data center situations. Finally, Chapter 8 presents somstandardizing storage management for UNIX and Windows-bas

The what and why of standardized stThe Veritas Storage Found


18/167

The what and why of standardized storage managementThe Veritas Storage Foundation by Symantec

18


19/167

The Storage Foundationvirtualization platform

This chapter includes the following topics: The Veritas Storage Foundation

The VxFS file system

Software integrated with the Storage Foundation

Whats new in the Storage Foundation

The Veritas Storage FoundationThe Veritas Storage Foundation core consists of a file system, caintegrated with a volume manager, or storage virtualizer, calledFoundation software is available for major UNIX server platforSolaris, HP-UX, AIX, and Linux (Red Hat and SuSE). The Storaghigh-performance, scalable toolset formanaging enterprise stopresents a common user interface on all major UNIX platformsscalability, and the common management interface combine wifunctions, such as dynamic management of storage network I/Othe Storage Foundations two primary value propositions:

The lowest cost mechanism for basic management of divers

Chapter


20/167

Components of the Storage FoundationFigure 2-1 illustrates the position of the Storage Foundation in the ssoftware components between business applications and data.

Figure 2-1 The Storage Foundation in the data access software

Files

Virtual volumes

Applications & databases

File System

Volume ManagerLogical units

(LUNs)

As Figure 2-1 suggests, the VxVM volume manager controls access tstorage connected to servers either directly, as is typical with disk dlogical units (LUNs) presented by disk arrays over a storage networkthrough adapter and device drivers that are part of the hosting operVxVM organizes the block storage of these devices and presents it uform of disk drive-like virtual volumes.

Virtual volumes

The Storage Foundation virtualization platformThe Veritas Storage Foundation

20


21/167

The disk drive-like behavior of volumes provides a powerful advall storage utility programs, file systems, and database managedesigned to manipulate data stored on disk drives. By acting likvolumes become usable by these applications and data manageincremental effort required to adapt them. Applications can takthe superior data reliability and I/O performance delivered by vwithout being specially modified to do so.

Applications and data managers like the VxFS file system that aadvanced functional properties of volumes can take advantage example, on administrative command, VxFS can increase the sia file system occupies, and modify file system data structures tadditional storage capacity.

Applications developed by other vendors can take advantage ofproperties as well. For example, the Oracle Disk Manager (ODM)VxVM to read data from a particular mirror in a mirrored volumwrite logging for mirrored volumes because ODM assumes contresilvering after a crash.

Volumes and LUNs

In many respects, the logical units (LUNs) of storage presented bsimilar to volumes. Both behave as disk drives from the applica

manager point of view. Both are implemented by software desigmultiple disk drives and present upward a disk drive-like virtuaenhance I/Operformanceand data reliabilityover what is availathat comprise them. The LUNs presented by disk arrays differ froin the six important respects listed in Table 2-1, however.

Table 2-1 Differences between LUNs and VxVM volumes

Differences between VxVM volumes and diskPropertyThe software that presents LUNs to a storage netwarray, so its scope of control is limited to disk driveVxVM runs in application or database servers, so iany LUNs that are accessible by a server into a virif the LUNs are presented by different disk arrays

Scope ofapplicability

The Storage Foundation virThe Veritas S


22/167

Table 2-1 Differences between LUNs and VxVM volumes (conti

Differences between VxVM volumes and disk arraProperty

Configurationoptionsfor diskarray LUNs tendto be limto those supported by VxVM. For example, VxVM placelimit on the number of mirrors (identical copies)or stripof data on different devices, whereas most disk arrays asmall numbers of both.

Virtual devicetypes

Operating systems typically name disk drives and disk according to theirI/Obus or storagenetwork addresses; amust usually maintain lists that correlate devices with andfile systems.VxVMprovides a flexiblenamingschemnames that relate objects to each other, as well as compfor administrators to name volumes according to applicrequirements such as their physical locations, the file sdatabases they contain, or other business purpose.

Virtual DeviceNaming

Most disk arrays require that all disk drives that are paLUN be of the same type. VxVM can support volumes cdifferent types of storage devices. For example, a LUN pan enterprise disk array can be mirrored with another pa different array of thesame type,with a LUN presentedtype of array, or even with a disk drive connected directcontrolling server.

Configurationssupported

Disk drives in an array are typically connected to their one or two physical I/O paths. In contrast, disk array LUconnected to servers by two or more storage network pamanages multiplepathsbetween LUNs andhostsfor faiand for diskarraysthat support concurrent multipathacan improve performance as well.

Multipath access

Costs and benefits of virtual volumesThe benefits of server-based volumes do not come without a cost. Wcomputing power to manage disk array LUNs is essentially a sunkit is provided by the disk arrays processor, VxVM uses application scomputing power to manage the volumes it presents For most volu

The Storage Foundation virtualization platformThe Veritas Storage Foundation

22


23/167

The VxFS file systemThe second component of the Storage Foundation core is the VxVxFS is an advanced-function, high-performance POSIX- compfor major UNIX platforms and Linux. POSIX compliance meansapplications can take advantage of VxFS benefits with no modifadaptation. Overfifteen years of development and deployment imade VxFS the most robust, scalable, highly-functional file sys

the UNIX and Linux environments.

Software integrated with the Storage FoundThe value of the Storage Foundation core lies not only in featurebut also in integration with other major data center software cofrom Symantec and from other vendors. Figure 2-2 represents g

Table 2-2 describes the major enterprise software components Storage Foundation integrates to provide comprehensive storagmanagement capabilities across the data center and across the

The Storage Foundation virT


24/167

Figure 2-2 The Storage Foundation and related Symantec softw

Volume

Virtual volumes

Logical units(LUNs)

Files

DatabaseEditions

Storage FoundationManagement Server

Comm

File System

ApplicationsDatabase managers

Table 2-2 Components integrated with the Storage Foundation

DescriptionComponent

Manages and reports on Storage Area Network (Scomponents, including disk arrays and network s

directors, from a single location (management co

CommandCentral

Manages and reports on Storage Foundation objeVxVM disk groups, volumes, and access paths, Vxfile systems, and groups of these objects, from a sIn addition, consolidates the basic tasks requiredcertain higher level services suchas migration of

Storage FoundationManagement Server

The Storage Foundation virtualization platformSoftware integrated with the Storage Foundation

24


25/167

Table 2-2 Components integrated with the Storage Found(continued)

DescriptionComponent

Provide customized management of and repFoundation objects specially adapted to majmanagement systems, including Oracle, DB2Database Editions encapsulate storage and f

tasks common to database environments soperformed by database administrators (DBArecourse to system or storage administrator

Database Editions

Manages applications and the resourcestheyincluding local and wide-area failover for rasystem failures and site disasters respective

Cluster Server

Working in concert with these components, the Storage Foundacomprehensive solution to datacenterand enterprise storage andproblems; a solution whose capabilities evolve to meet the growenterprise computing as it grows in scope and becomes increas

Whats new in the Storage Foundation

The first versions of the VxVM volume manager and VxFS file sthe core of the Storage Foundation were delivered in the early 1fifteen-plus intervening years, the nature of UNIX-based enterphas changed significantly. UNIX systems have become more poworders of magnitude, both in terms of processing ability, and inmulti-gigabyte primary memories that enable them to cope witdata-intensive problems. As enterprises have become comfortabcritical applications to UNIX-based systems, the number of UNI

centers has grown substantially, creating more focal points for management, and in addition, introducing the challenge of manrelationships between applications and the computers they run

Perhaps even more significantly, online storage has moved outis connected to the computer systems it serves by storage netw

The Storage Foundation virWhats new in the S


26/167

compatibility with users already-installedconfigurations.The most rFoundation version, Version 5, improves performance and providesnew functions, as well as responding to todays three most significaenterprise storage and file management:

A central view and point of control for all data center storage resprovided by the Storage Foundation Management Server

Cost-effective utilization of two or more cost-performance tiers o

center storage, provided by the Dynamic Storage Tiering facility Integrated virtualization of network storageacross entire data cen

by the Storage Foundation Volume Server

The sections that followdescribe theserequirements and howStoragVersion 5 meets them.

Centralized storage and data managementAs the number of systems and storage devices in an enterprise datasimply keeping track of what installed in a data center becomes a sichallenge. With enterprise disk arrays capable of presenting thousato hundreds of hosts on dozens of ports, answering seemingly simpbecomes a significant challenge:

Which hosts are using LUNs presented by a given array?

Which arrays are providing storage to a given host?

Which VxVM disk groups exist in the data center and which hosthem?

Which storage is VxVM protecting by mirroring, RAID5, or snap

Which file systems are mounted by which hosts and how well-utthey?

Where in the data center is there underutilized storage that couredeployed?

What storage and file system-related alerts have been raised acrcenter?

The Storage Foundation virtualization platformWhats new in the Storage Foundation

26


27/167

the answers. But as the number of systems and storage devicesgrows, storage management by spreadsheet is no longer effectivadministrators, each responsible for different applications anddifferent servers,must interact with storage administrators chathe storage needs of their applications. Storage administrators data center-wide view of how storage is allocated and how it is bwell as a central point from which they canperformcommon stotasks such as the transfer of VxVM disk groups from one host t

migration of data from one disk array to another.

Beginning with Version 5, the Storage Foundation Managemencapability providesadministrators withglobal views of and compabout all Storage Foundation objects. Moreover, SFMS makes it pand control common storage and file system management taskconsole, wherever they are executed. SFMS discovers Storage F(disks and disk groups, volumes, file systems), and related objects

and managed servers, and populates a central database with thstates. SFMS components called agents run on systems with thFoundation installed and supply information withwhich SFMSudatabase as changes in the storage and data environment occur

Storage Foundation Management Server includes reporting andcomponents that can be used to track the state of Storage Foundrespond to storage-related events, as well as active managementt

the creation and execution of common storage management-reSFMS reports can:

Summarize overall data center storage status (for example, aof storage allocated to file systems across the data center)

Identify candidates for management action (for example, sylowest utilization of the storage assigned to them, which micandidates for redeployment)

Locate problem areas (for example, volumes that are stoppestorage access paths that are interrupted or at risk of interradditional component fails

SFMS alert tracking provides an overall picture of storage-relat



28/167

by busy administrators under pressure to perform tasks with whichonly marginally familiar.

In addition, recognizing that every data centers procedures are unimakes it possible for users to store custom storage management tasand execute them on demand, tracking their progress through the SThus, for example, an administrator could define a SFMS task consicommands to defragment the web page file systems on all of the entservers. The task definition could be saved in the SFMS database foon demand, for example whenever some external event such as a wewas likely to have caused significant change across the web servers

Conclusion: Storage Foundation Management Server provides a sinpoint from which administrators view all of a data centers storage vdevices and file systems, and can manage them with full visibility oconsequences of actions on the entire data center storage complem

Effective utilization of multi-tier storage

Multi-tier storagetheblending of virtual or physical storage devices wI/O performance, data availability, and relative cost characteristics tdifferentiatedonlinestorage forcomputer systemsis increasingly poenterprise IT organizations as a way of minimizing the average coststorage while continuing to meet user demands.

Tiers of storage can be defined by: Type of hardware, for example, low-cost disk drives directly atta

vs. LUNs presented by an enterprise disk array

Age of hardware, for example, 36 gigabyte disk drives vs. 73 gigadrives in the same enterprise disk array

Configuration, for example, RAID5 virtual devices vs. striped an

virtual devices of the same capacityAll of these options deliver unique availability and performance chaat an associated cost. Virtualization technology allows the use of mtechnologies in combination. For example, RAID LUNs presented byarrays are sometimes mirrored by VxVM or similar host-based volum


28


29/167

Enterprises segregatedatasetsforother reasons as well. For examdesirable to separate transactional data from large data streamreasons, because the access patterns of the two are so differentdata sets that are both accessed at the same time by different apreside on different storage devices to minimize I/O resource conand similar cases, there is benefit to be gained from segregatingand storing different types of data on devices in a specific tier, ecapabilities are identical.

Fundamentally, enterprises organize their digital information iclosely associated with business purposes. Documents, transactiaudio and video tracks, and so forth, are all conveniently represeach with a business purpose and value. This makes the file anoptimizing storage performance, availability, and cost.

To exploit multiple storage tiers, an enterprise would place eachon a type of device whose cost was commensurate with its busi

critical files would be placed on more expensive, but higher-perfreliable devices; less critical files would be placed on less costlypresumably lower performance and reliability.

Matching one file, or a group of files, to the appropriate storagedifficult. Administrators can force files to be placed appropriateby assigning users or applications to file systems formatted on types of devices. The challenge in utilizing multi-tier storage ef

the numbersplacing millions of files on the right devices is fato perform effectively without some form of automation.

Moreover, the appropriate type of storage for a given file changa file ages, is accessed more or less frequently, grows or shrinkswithin its file system logical name space, the right type of storagFor example, newly created transaction records are typically acas orders are placed, scheduled for shipment, billed, and inquir

records age, access is less frequent, but they must neverthelessoccasional inquiries, monthly closings, data mining, and so forcost of keeping transaction records onlinecan be reduced substaolder records to lower performing, less expensive storage deviceare accessed less frequently.



30/167

The ability to distribute a single file system (name space) acroVxVM volumes with different properties

Automatic policy-based placementand relocation of files amongavolume set

The individual volumes used by a VxFS multi-volume file system areto applications. But VxFS itself remains aware of each volumes tier cand implements pre-defined policies that dynamically control the t

on which each file is stored.VxFS administrators define policies that automatically control bothlocation and the circumstances under which VxFS relocates existingconsist of rules that restrict files locations to designated storage tierules cause files to be created and extended within specified tiers, arelocated to other tiers when they meet certain qualifications such activity, and location within the name space.

Conclusion:Storage Foundation utilizes multiple tiers of online storaby automatically creating each file on the right type of storage, anto its new right type of storage when conditions around it changeadministrative intervention.

Virtualization of storage across the storage network

Starting with the first RAID-capable disk arrays of the early 1990s, presented to enterprise applications has increasingly become virtuamost file systems and database management systems store their daton disk drives, but on disk drive-like virtual devices. The virtual devof one or more disk drives plus a layer of software that coordinates activities and presents disk drive-like virtual devices to file systemsmanagement systems, and storage utilities.

Virtual storage devices are functionally equivalent to disk drives as f

and writing data are concerned, but they offer superior I/O performavailability, data reliability, and function compared to the disk drivethey consist. As storage has moved out of the server and onto the nevirtualization technology has evolved to encompass logical connectivdevices to systems as well.


30


31/167

the storage it virtualizes. Table 2-3 summarizes the scope of virdisk arrays, storage network switches, and servers.

Table 2-3 Scope of storage virtualization

PropertiesVirtualization location

Disk drives: Can virtualize disk drives whas no ability to control drives in other

Hosts: Can present virtual devices to hoand logical connectivity to the arrays SA

Disk array firmware

Disk drives: Can virtualize devices preseThese may include both disk drives andalready been virtualized by disk array fi

Hosts: Can present its virtual devices toand logical connectivity to its SAN ports

Storage network switchfirmware

Disk drives: Can virtualize disk drives anstorage network ports it has physical anconnectivity through the storage networ

Hosts: Can present virtual devices to appon the application or database server, buapplications on other servers.

Application or databaseserver software like VxVM

Studying Table 2-3, it becomes obvious that none of the convenstorage virtualization operates on a global scale, that is, none carbitrary pool of a data centers storage devices and present virarbitrary collection of computers. In data centers with only a festorage either directly attached or configured statically, the limvirtualization is of minor concern. But as enterprises deploy lasmaller blade servers, and storage redeployments become a dai

lack of a larger vantage point from which to manage storage incthe ability to utilize storage assets effectively.

Responding to this need for a larger scope of control over data cVersion 5 of the Storage Foundation enhances volume managerprovide centrally controlled virtualization of pools of storage re



32/167

Administrators manage storage resource pools for groups of hosts fpoint, with visibility to all resources and storage consumers.

Conclusion: With the Storage Foundation VolumeServer facility, stosharing and serial data sharing between arbitrary groups of UNIX hsupported types becomes fast and seamless.


32


33/167

A closer look at the StorageFoundation core: the VxVMVolume Manager

This chapter includes the following topics:

Why Storage Foundation?

The VxVM volume manager

New in the Storage Foundation: multi-host coordinated stor

Why Storage Foundation?The core components of the Storage Foundation, the VxVM voluthe VxFS file system, occupy a position at the heart of the data stack. So fundamental are host-based storage virtualization andand management to enterprise computing that UNIXand Linux o

include native virtualizers and file systems. This raises the queinvestment in the Storage Foundation is justifiable, or, put anotUNIX and Linux native file systems and volume managers are gmeet basic enterprise needs.

For most enterprises the answer to this question is a strong aff

Chapter


34/167

Platform support that makes it possible to manage storage and da data centers UNIX and Linux platforms with a single set of procontrolled from a single management console.

Integration that makes Storage Foundation I/O performance, datand storageasset management easily accessibleto othersoftwareicomponents such as cluster and storage network managers, as wdatabase management systems and applications.

A more detailed understanding of how the Storage Foundation delivrequires a closer look at both the nature of storage virtualization ancore Storage Foundation components implement their functions.

Storage virtualization

In information technology, virtualization usually means combiningavailable physical components to produce the behavior of a compon

that is difficult (or impossible) to obtain. For example, disks with 1,0hour mean time to data loss and disks capable of executing 2,500 I/Osecond are impossible to build, or at least prohibitively expensive tovirtual disks with these properties can be created easily by mirrorindata on two or more disks and striping block addresses across a dozdisks respectively.

Storage devices are virtualized by software that manages one or mo

devices, mapping client read and write requests to virtual storage toto the managed devices. Virtualization has two parts:

Algorithms that determine how client I/O requests map to I/O opdisks or disk array LUNs.

Configuration information that drives the algorithms. Configurainformation defines which physical devices comprise which partdevices, and the types of algorithms used to access them.

For example, the algorithm for mirrored virtual devices is to executewrite request on all devices and to choose one device to execute eachrequest. For striped virtual devices, the algorithm converts virtual daddresses into block addresses on oneor more physical devices and exread and write requests by issuing commands to the selected device

A closer look at the Storage Foundation core: the VxVM Volume ManagerWhy Storage Foundation?

34


35/167

of which virtual devices so that it can execute its algorithms comust virtual storage configuration information persist across smust persist across storage device failure and replacement. In astorage configuration information is the most valuable data an is the key by which application data is located. If virtual storaginformation is lost, an enterprise is in the frustrating position oits important data is physically intact on functioning devices thcontrol, but being utterly unable to locate and access it.

The VxVM volume managerThe VxVM volume manager is a host-based storage virtualizatiSolaris, HP-UX, AIX, Red Hat Linux, and SuSe Linux platforms.above the hardware device driver layer in the data access softwFigure 3-1 illustrates. Disk and I/O adapter drivers are typically

of an operating system, or by the vendors of storage and I/O adBecause it is essentiallya client of any and alldisk storage driversin a system, VxVM can virtualize any disk storage device that ansupports. For virtualization purposes, VxVM treats all disk drivLUNs presented to it through drivers identically as generic VxV

The ability to virtualize all types of storage is a key advantage icenter storage strategy independent of hardware vendors. Whe

devices are introduced, or oldones are retired, no change in virtuis required. If the new devices are roughly equivalent to the oldI/O performance and data protection, they can be substituted foand instantly recognized and managed by VxVM.

A closer look at the Storage Foundation core: the VxVThe VxV


36/167

Figure 3-1VxVMs position in the data access software stack

Device driversD

Disks

Volume Manager

Logical units(LUNs)

File System

Virtual volumes

VxVM transactionsVxVM treats all disk drives and disk array LUNs visible to it throughstack identically as abstract disk objects, which it organizes into disVxVM virtualizationand other management operations occur withinof a disk group. For example, a disk in one disk group is never mirrodisk in another group.

Thedisk group organization makes storagemanagementmore reliablefor large numbers of storage devices in complex networks connectedhosts. For example,to transfer control of a set of disks from one applito another, an administrator would split the disks from their disk grseparate group, deport the new group from the controlling server, aon the target server If anything were to go wrong during this seque

A closer look at the Storage Foundation core: the VxVM Volume ManagerThe VxVM volume manager

36

A l l k t th St F d ti th V V


37/167

Atomicity, meaning that at the end of a transaction, either aoperations that comprise it have executed to completion, orreversed so that the effect on the system is as though the trnever occurred

Consistency, meaning that a transaction transforms a systemstate to another (primarily a property of application correct

Isolation, meaning that no matter how many transactions a

concurrently, each behaves as though it were the only one rsystem

Durability, meaning that when a transaction signals that it iresults persist, even if the system on which it is running faiafter the completion signal is given

VxVM transactions manage disks, disk configurations, and diskFor example, allocating part of a disks capacity for a volume, com

into a mirrored volume, and adding a disk to a striped volume aas transactions. Because VxVM management operations are traadministrators need not concern themselves with recovering froccur in mid-operation; VxVM makes sure that management opcomplete or are backed outdisks and other objects are never lthe apparent control of two servers.

Private regionsVxVM disk groups are self-describing, meaning that the disks ininformation about group membership, volumes, and other VxVFigure 3-2 illustrates, VxVM divides the total storage capacity omanages into a small (a few megabytes) private region in whichmetadata, and a much larger data storage region that it makes aapplication use.


A closer look at the Storage Foundation core: the VxVM Volume Manager38


38/167

Figure 3-2 VxVM private and public regions

Device drivers D

Disks

Volume Manager

Logical units(LUNs)

Privateregion

Datastorage

Taken together, the private regions of the disks in a disk group desccomplete configuration of the group. Each disks private region contfacts about the disk group as a whole, a description of the volumes aVxVM objects of which the disk is a part, and redundant copies of inabout other disks in the group. All information about a disk group is

or more disks private regions so that the durability property of VxVMcan be guaranteedeven if one or more disks fails, VxVM can use mthe private regions of the remaining disks in the group to reconstruvolume configuration.

VxVM achieves the remaining transactional properties, atomicity, cand isolation through a combination of carefully sequenced private reand its own resource locking mechanism. The result is a repository

storage configuration information that is robust, both in the sense oagainst multiple storage device failures and in the sense of being upa way that it can be recovered to a correct starting point if a systempoint during an update. Because VxVM disks are self-describing, coinformation travels with them, so that, for example, when control of

i i i d i d l i f d f


38

A closer look at the Storage Foundation core: the VxV


39/167

VxVM object namingUNIX systems discover the storage devices connected to them wby probing I/O bus addresses, or by querying storage network ndiscovered device is given a name that is related to the I/O bus it is discovered, or to its worldwide storage network name. The part of the operating systems device tree,and are used in managsuch as virtual volume and file system creation.

While operating system device naming schemes guarantee thatvisible to a system is named uniquely, the names they produce shortcomings:

They are difficult for human administrators to use on a dailthey have no relationship to business purpose.

They fail to distinguish between separate devices and a singor more connections to the system.

VxVMs user-friendly device naming capability overcomes the fshortcomings. The second is overcome by the Dynamic Multi-pcapability discussed in Chapter 5.

For a system with only a few storage devices connected to it, desimple. Storage device configurations change infrequently, so anaturally come to memorize device names and their relationshof the IT operation during the course of their daily work. It is e

that the disk or LUN called /dev/dsk/c0t0d1s0 contains the fiaccounts payable application.But as systems are reconfigured,anand more disk arrays are installed in a data center, relationshipquickly becomes impossible to hold the storage configuration inmemory. As a data center acquires more administrators, the procompounded because the administrators must coordinate theirthat each administrators mental picture of the data centers sto

as all others.Storage virtualization compounds the difficulty further by addobjectsvirtual volumes and the storage components that compnames must be specified correctly in management operations.

VxVMs object naming scheme mitigates the problem of quick




40/167

Enclosure-based naming of VxVM disk objects

The first two columns of the display in Dialog 3-1 which designate tobject type and object name respectively, illustrate the first two of taspects of the VxVM object naming scheme.

Dialog 3-1 VxVM object naming

[1] # vxassist -g testdg make appvol 100m nmirror=2

[2] # vxprint

[3] Disk group: testdg

[4] TY NAME ASSOC KSTATE LENGTH PLOFFS STATE TUTIL0 PUTIL0

[5] dg testdg testdg - - - - - -

[6] dm c3t1d0s2 c3t1d0s2 - 71680896 - - - -

[7] dm c3t2d0s2 c3t2d0s2 - 71680896 - - - -

[8] v appvol fsgen ENABLED 204800 - ACTIVE - -

[9] pl appvol-01 appvol ENABLED 204800 - ACTIVE - -

[10] sd c3t1d0s2-01 appvol-01 ENABLED 204800 0 - - -

[11] pl appvol-02 appvol ENABLED 204800 - ACTIVE - -

[12] sd c3t2d0s2-01 appvol-02 ENABLED 204800 0 - - -

Lines [6]-[7] in Dialog 1 represent two dm (disk media) objects. The

drives or disk array LUNs that have been placed under VxVM controby which thedisks are addressed in VxVM administrative commandsc3t1d0s2, appear in the second column of the display. The third col

display represents an object with which the VxVM object is associatcase, the operating systems name for the device. VxVM has set the

which it recognizes the devices to be identical to those assigned by tsystem. If for some administrative reason it were more appropriate tdisk objects other names, that is also possible.

Line [8] represents a VxVM virtual volume (object type v) that is pr

clients.The vxassist commandin line [1] of theDialog creates thev

is mirrored, and names it appvol for administrative convenience. A V

consists of one or more objects called plexes (object type pl), which in

of subdisks (object type sd). Lines [9] and [11] illustrate that the pwhich are created automatically by VxVM during the course of voluare named appvol-01 and appvol-02, and are thus immediately ide

being associated with the appvol volume. Lines [10] and [12] repre

objects c3t1d0s2-01 and c3t2d0s2-01 which VxVM names by defa


40



41/167

houses them (SHARK0_26 and SHARK0_27 in Dialog 3-2). Enclosu

are generated by array model-specific VxVM modules called ArModules (ASMs).

With enclosure-based naming, physical locations of LUNs are tidentifiable. If application or business oriented names are more acan substitute a user-defined enclosure name for that reporteddisks are named according to the substitute name.

Dialog 3-2 VxVM enclosure-based naming

[1] Disk group: tcrundg

[2] dg tcrundg default default 16000 140131242.57.vm280r1.veritas.c

[3] dm IBM_SHARK0_26 IBM_SHARK0_26 auto 65536 1883008 -

[4] dm IBM_SHARK0_27 IBM_SHARK0_27 auto 2048 1946496 -

[5] v appvol - ENABLED ACTIVE 20480 SELECT appvol-01 g

[6] pl appvol-01 appvol ENABLED ACTIVE 20480 STRIPE 2/128 RW

[7] sd IBM_SHARK0_26-01 appvol-01 IBM_SHARK0_26 0 10240 0/0 IBM_SHARK0_

[8] sd IBM_SHARK0_27-01 appvol-01 IBM_SHARK0_27 0 10240 1/0 IBM_SHARK0_

As these simple examples illustrate, VxVM automates default othat related objects can easily be correlated with each other. Ennames automatically correlate LUNs with the disk array enclosthem. Moreover, VxVM object names are completely at adminisAdministrators can design storage naming schemes to meet enrequirements, be they application-oriented, data center layout-

designed for some other business purpose.

Multipath access to storage

As data centers have evolved from configuring storage for eachand adopted the pooled storage concept enabled by storage netwnetwork itself has become an important factor in delivering higapplication availability. Not only must virtual storage devices bability to access the physical devices that comprise themmust benetwork failures as well.

The primary mechanism for making connections between hostdevices failure-tolerant is to provide multiple paths between th

gThe VxV



42/167

of the access paths, and tracks outstanding requests until their execcomplete.

The most failure tolerant form of multipath access is to configure twseparate physical networks, or fabrics, and connect both host servedevice to both fabrics, as Figure 3-3 illustrates.

Figure 3-3 Multiple paths to networked storage devices

Disk array

Path 1

Controller

Path 2

Mirrored LUNspresented by

disk arraycontroller

Port

Switch

Controller

Port

Switch

HBA1 HBA2

Host server

In block diagram form, Figure 3-3 represents the primary physical c

in the path between server and storage devices (in this case, mirrorevirtualized by disk array firmware.

The two blocks labeled switch in Figure 3-3 represent separate stor

fabrics with no shared components except the end point host servert S t h t b d t (HBA ) bl di




43/167

The host servers I/O software stack is able to re-channel reqfailed path to the alternate path so that applications accessuninterrupted.

VxVMs Dynamic Multi-pathing (DMP) featurefulfills both of thDMP builds on basic operating system device discovery to recogpaths to the same disk drive or disk array LUN. If the path to a dreason, DMP re-routes requests to an alternate path. In most catransparent to applications. For so-called active-passive disk ar

specific action to communicate with a device on an alternate pathe required actions.

Figure 3-4 Dynamic multi-pathing in the storage I/O stack

Disk c1t0...Devices asdiscovered byoperating system

V-Disk c1t0...

Disk c0t0...

DMP feature

LUN A

Device driver

Path 0 Path 1

Volume Manager

Figure 3-4 illustrates that DMP occupies a position below the Vxlayer in the I/O stack. During its discovery phase, which occursand may also occur on demand, DMPqueries similar devices to dthey are actually different devices, or the same device presentesystem through twohostbusadapters.For each device to which i

The VxV

A closer look at the Storage Foundation core: the VxVM Volume ManagerTh V VM l

44


44/167

If a path to an I/O device fails, DMP simply re-routes outstanding anrequests to another path, performing any disk array-specific actioneffect path failover.

A secondary benefit of DMP is the ability to use two or more paths tconcurrently to increase throughput. This feature is especially valuenterprise-class disk arrays with large cache memories. For certain applications, sizable percentages of I/O requests are satisfied by refarray cache. Because there are no seeking or rotational delays for th

data transfer is the factor that limits I/O performance. If a disk arraconcurrent access to its LUNs on two or more paths, DMP can routeI/O request to the least busy path for improved overall throughput. Dseveral algorithms for balancing I/O load across multiple paths.

The primary benefit of DMP is an increase in data availability at thebecause it makes access to storage devices tolerant of I/O path failurebenefit is improved I/O performance when used with disk arrays tha

concurrent access on multiple I/O paths to the LUNs they present.

Online reconfiguration

As data centers grow in size and complexity, and as the velocity of cdigital data processing increases, server and storage reconfigurationfact of daily life. Application data outgrow allocated storage capacitvalue to their owners increase, requiring greater protection; systemexcess of storage capacity donate part of it to other systems with to

Whatever the reason, reconfiguration spells downtime for applicathat is increasingly becoming an anathema to administrators requirtheir applications up and running around the clock. Taking a large fdatabase out of service so it can be backed up and restored to a largeoften intolerable from a business standpoint. Even the time requiredata set from non-redundant devices directly to mirrored ones may

operations adversely.

VxVMs architecture eliminates or at least mitigates many of these sreconfiguration issues by making it possible to perform them whileand in use by applications. Volumes can be resized, both to greater ca

li ti d it d t l it h d t




45/167

VxVM can relocate a single mirror or column of a striped volumor LUN to another. This feature is useful when potential disk fasuspecteddata can be moved from the failing disk to an alternin use. It is also useful for migrating data from older devices to

VxVM performs all of these reconfiguration operations in the bI/O capacity that is not being used by applications. But all data operations entail significant amounts of I/O, so it is prudent to speriods when production applications are not at peak load. Wh

reconfiguration does accomplish is the elimination of downtimcan be fully functional while reconfiguration is going on, even iperformance. Thus, online reconfiguration turns a hard boundaInstead of requiring that an application be out of service for as reconfigure its storage, VxVM only requires that reconfiguratioa time when any potential performance degradation can be tolebusiness standpoint.

Snapshots

Applications are increasingly required to operate around the cloadopters like air traffic control, to business-to-business functiotransfer, to web retail applications, what started as a convenienimperative. An information service that isnt available around tnot viable. Somewhere, there is a competing service that is conti

But at the same time, enterprises are becoming increasingly dedigital information to operate. It is a rare enterprise, no mattercan function without its computer systems and digital informaenterprises must preserve and protect their digital data, just asprotect cash, capital equipment, and other valuable assets.

These two trends conflict with each other. Onethe onehand,datafor use around the clock. On the other, it must remain unmodif

for copies to be made to protect it against destruction.

VxVM snapshots that capture split-second images of large dataphotographic snapshots capture images of physical action, helpconflict. A VxVM snapshot of a large data set can be backed up

d hil th d t it lf ti t b d b

The VxV


46


46/167

the disadvantage that initial creation takes a long time, because evedata set whose snapshot is being taken must be copied to the snaps

Space-saving snapshots are virtual snapshots of volumes. Physicallspace-saving snapshot includes only the prior images of data that isafter snapshot creation. Storage capacity for the prior images is allothey are updated by applications. Space-saving snapshots are presecompletedata sets, but they satisfy read requests for unmodified datto the original image. For this reason, they cannot be moved off-hos

computer for auxiliary processing.

Space-saving snapshots use storage capacity very economically, esplarge data sets with low rates of changethey consume storage capaproportion to the amount of data modified during their lifetimes raproportion to the size of the data set. Thus, it is feasible to take freqspace-saving snapshots so that data sets can be quickly restored to f(recovery points) if they become corrupted.

The VxVM FastResync feature greatly enhances the usability of boand space-saving snapshots in repetitive applications and for recovapplication-caused data corruption. VxVM maintains Data Change Lthat track changes both for active snapshots and for thevolumes whothey represent. Using DCLs as guides, VxVM can quickly restore a cvolume to a prior state by overwriting changed blocks with the contprior images from a snapshot. Similarly, a snapshot whose useful li

be updated to a more recent one by overwriting changed blocks fromvolume to it. Thus, for example, a full-copy snapshot can be taken obackup, and after the backup is complete, be brought back on-host ucurrent state of its original volume by copying changed data to it, anoff-host again for the next backup cycle.

VxVM snapshots can dramatically increaseinformation service uptimit possible to capture instantaneous states of large data sets that caup or analyzed while production applications continue to process livSnapshots can be used to recover from corruption of production datadata to a state it was in at some point prior to the occurrence of the FastResync technology increases the utility of snapshots by reducinof copying required for both the restoration of prior volume states a

f hi f h t t th i i ibl




47/167

from physical devices and placing that virtual capacity under tor more application or database servers.

In a data center in which servers use directly attached storage (provisioning is relatively simple. Each storage device is permanwith one server. The only provisioning action required is to condevices appropriatelyto organize disk drives or disk array LUNstriped, or other types of virtual volumes.

When storage devices are connected to a network and accessiblhowever, storage provisioning becomes more complex. From a pdisk drives and LUNs, an administrator must select appropriateplace them under control of one server or cluster, and configurethat can be used by applications. In a data center with dozens oservers, and potentially thousands of disk drives and LUNs, keepphysical devices are available for provisioning and how allocateconfigured becomes difficult if not impossible to do in a timely

manner.Over its history, VxVMs storage provisioning capabilities have ewith the complexity and connectivity of the data center environit is used. VxVMs provisioning capabilities can best be appreciaunderstanding of the internal structure of VxVM volumes. Figuthe primary VxVM objects and how they are used to construct ththat are presented to applications.

Figure 3-5 VxVM objects

LUN

Disk array LUN VxVM disk VxVM subdisk VxVM plex

Privateregion

Available fordata storage

Privateregion

Other subdisks Other

Starting from the left side ofFigure 3-5, VxVM recognizes the L

The VxV


48


48/167

The public region of a VxVM disk can be subdivided into ranges of cblock addresses called subdisks. Subdisks are the underpinning thapossible to divide the capacity of a very large disk into smaller unitspresentation to applications as volumes, or alternatively, to distributof a single volume across multiple physicalresources to improveI/O p

One or more subdisks can be organized into a single block address rplex. While it is presented upward as a single block address range, thcomprise a plex may be distributed across a number of disk array LU

drives.

Finally, one or more plexes can be organized into a volume, which isvirtual storage on which file systems are formatted and presented foapplications.

This four-level object hierarchy of disk, subdisk, plex, and volume padministrator with precise control over storage resource allocation. an administrator can define subdisks using capacity on four differen

representing pairs of LUNs presented by two different disk arrays. Tof subdisks can be made into striped plexes, which in turn can be meach other and presented as a volume. Figure 3-6 illustrates this con

Figure 3-6 Constructing a VxVM volume: an example

Disks Subdisks Striped plexes

Mirroring

LUNs

LUNs

g



49/167

The resulting volume has two important desirable characterist

Striping tends to balance I/O load across the two subdisks o

Mirroring protects data on the volume against loss due to thone of the four disks that contribute capacity to it.

Subdisks, plexes, and volumes can all be constructed using the

console commands or the Veritas Enterprise Administrator graThis is referred to as the advanced approach to volume manage

advanced approach, a system administrator is responsible for dappropriate storage layout, for example, ensuring that the subdcomprise a striped plex are presented on different ports of the and that the two plexes consist of storage presented by differen

While working directly with VxVM subdisks and plexes gives adprecise control over volume configuration, the advanced approimportant shortcoming that becomes apparent as a data centerg

more complex. Administrators who manage each basic VxVM obkeep track of each subdisk and plex throughout its life, and, momust allocate every new or reconfigured volume so that it meetbusiness intent for data sets.

For example, if enterprisepolicy dictates that everycritical dataon volumes that are mirrored on LUNs presented by different dto the administrator to identify and configure appropriate LUN

require a new volume. Similarly, if enterprise policy dictates thamust be accessibleon at least twoI/Opaths, it is the administratto ensure thateverynew volume is configured from VxVM diskson at least two paths.

With disk arrays capable of presenting a thousand or more LUNof individual disks, subdisks and plexes is clearly impractical. Mindividually configuring thousandsof volumes from their basic Vis almost guaranteed to result in configuration errors, or at a menterprise data management policies not being followed.

Recognizing that data center growth, and in particular, online swas routinely creating situations in which the sheer amount ofmade management difficult and error-prone, Veritas introduce


50


50/167

It promotes configuration consistency by storing common install

parameters such as stripe width and mirror count in a file, and udefault values when executing management commands.

It relieves administrators of the responsibility of tracking disk uchoosing the disks from which to create, expand, or reconfigure wherever possible.

Thus, for example, creating the mirrored striped volume shown in Fusing the advanced configuration technique would require that an achoose two pairs of LUNs presented by the two disk arrays and reprVxVM disks, and issue vxmake commands to create the subdisks, plexe

the volume. Assuming that defaults have been set to create mirrorevolumes, using assisted volume configuration reduces the administcreate this volume to the single command shown in line [1] ofDial

Dialog 3-3 Using vxassist to create volumes

[1] # vxassist g diskgroupA make volume01 100g

[2] # vxassist g diskgroupA make volume02 100g diskgroupA01 dis

[3] # vxassist g diskgroupA make volume03 100g |diskgroupA03 |c

[4] # vxassist g diskgroupA make volume04 maxsize layout=raid5

The vxassist command simplifies volume creation to the greatest ex

while still providing the flexibility for administrators to configure stindividual requirements. For example, line [2] ofDialog 3-3 shows

administrator can specify the VxVM disks on which a volume shoulLine [3] illustrates how specific disks (diskgroupA03), and disks co

specific controllers (ctlr:c2) can be excluded from being part of a v

[4] shows how the default volume type (striped mirrored) can be ov

specifying layout=raid5, and also how the maximum available cap

disk groupcan be incorporated into the volume by specifying maxsiz

a numeric value. Dialog 3-3 illustrates basic examples ofvxassist;

options are available to meet the goals of simplicity and flexibility.

In all of the examples in Dialog 3-3, VxVM creates the basic subdiskobjects automatically, choosing appropriate locations for them withconstraints specified. For example, when the vxassist command is u



51/167

of using low-level VxVM commands to create VxVM objects bel

level.

This form of the vxassist command greatly simplifies virtual v

administration, and perhaps more importantly, reduces the potconfiguration errors. As disk arrays have become more complexusers have become more sophisticated about how they configurtheir online storage, the need for even more sophisticated virtuconfiguration has become apparent. For example:

Larger data centers often have two or more enterprise-classdefine policies that mirror critical data on LUNs in separate

The virtualization performed by disk arrays must be coordinwith VxVM virtualization. For example, mirrored LUNs presarray may not need to be mirrored again by VxVM.

As disk array capabilities havebecome more sophisticated, p

relationships such as that between LUNs and controller pornetwork are no longer valid.

As storage networks have grown and become more complexcomponents of automatically-configured volumes are reachmore independent paths has become a priority.

IT service providers that have service level agreements withorganizations often require that certain physical storage cap

only to certain users.To meet the need for more general forms of automated volume Version 4 of the Storage Foundation introduced the Intelligent Provisioningfacility. Intelligent Storage Provisioning makes it povirtual volumes based on templates that describe required propgeneral way. For example, a template may specify that mirroredmirrored either by VxVM or by the disk arrays that present the

disks. Alternatively, a template may specify that mirrored voluof mirrors in two different disk array enclosures.

Equally important, with Intelligent Storage Provisioning, arbitrpairs called attributes may be assigned to disks. These attributuser-determinedthey can be used to represent any desired bu


52


52/167

sales volumes, and accounting volumes, each specifying that it mus

from storage devices with the corresponding attribute.

Version 5 of the Storage Foundation links Intelligent Storage Provisassisted volume configuration. Beginning with Version 5, it is possibdisk attributes in vxassist commands, making it possible to constr

to be created from disks with specified attributes or conversely. Diaillustrates how user-defined configurations can be created and modand easily, while still preserving an enterprises intent for its storag

Dialog 3-4 Using vxassist combined with Intelligent Storage Pro

[1] # vxassist g diskgroupB make volumeB1 100g mirror=faultzo

[2] # vxassist g diskgroupB make volumeB2 100g department=acc

[3] # vxassist g diskgroupB make volumeB3 100g building=build

[4] # vxassist g diskgroupB mirror volumeB1

[5] # vxassist g diskgroupB growby volumeB1 100g

Line [1] ofDialog 3-4 illustrates the intrinsic property of separatiomirror=faultzone parameter indicates that the volume is to be mir

that VxVM is to construct the mirrors on disks with different valuesfaultzone attribute. Lines [2] and [3] illustrate the use of the user

parameters department and building to specify that volumes shou

from disks reserved for the accounting department or located in burespectively.

The real value of Intelligent Storage Provisioning, however, is that iadheres to user intent when creating a volume, but it preserves thatextending it. For example, the command in line [4] ofDialog 3-4 ad

mirror to volumeB1. Because volumeB1 was created using mirrors in

faultzones, VxVM will ensure that the third mirror added when lin

executed is in a separate faultzone as well. Similarly, when the volu

increased by the command in line [5], the intent that its mirrors be

faultzones is preserved as well.Combining the simplicity of the vxassist command with the gener

Intelligent Storage Provisioning simplifies virtual storage configurameeting and preserving enterprise intent for storage configurationsstorage net orks ith ad anced disk arra capabilities

A closer look at the Storage Foundation core: the VxVNew in the Storage Foundation: multi-host coordinated sto


53/167

be deported from a Solaris system and imported and used on an

Linux system. This feature of VxVM, known as portable data colays the groundwork for serial data sharingthe use of data creplatform by another UNIX platform of a different type. For examon a Solaris platform can be transferred to a HP-UX or AIX platgeneration or data analysis. Use of data on a platform differentwhich it was created also requires that both platforms utilize coand file data formats. VxFS file systems use a common file syst

platforms, and a format that is convertible between UNIX and Lapplications use platform-independent data formats; others, sumanagement systems, provide filters for transporting data fromanother.

As more and more applications and data managers respond to tdata they produce on different platforms, the ability to move vofiles and databases will be of increasing importance. VxVM prounderpinnings for transferring data between platforms withoulengthy network copy operations or low-performing NFS cross-systems.

New in the Storage Foundation: multi-hostcoordinated storage management

Since its inception, VxVM has been a host-based storage virtuaThe locus of control and management of virtual volumes has becomputer on which a VxVM instance is running. The scope oveinstance exercises control is the set of disk drives and LUNs thaby the host on which it is running.

But thehost-centric viewis becoming less and less adequateas dbecomes more complex and the parts that comprise it become m

Recognizing the growing need of enterprises to manage their vacross a broader front than the individual server, Symantec hasevolution of virtual storage management from host-centric to dwith the introduction of the Storage Foundation Volume Server

To understand the volume server it is helpful to review the functi

A closer look at the Storage Foundation core: the VxVM Volume ManagerNew in the Storage Foundation: multi-host coordinated storage management

54


54/167

Figure 3-7 Structure of host-centric VxVM storage virtualization

EventsVxVM

Application control

Carp

c

Control(vxconfigd)

I/O path

Managementactions

Controlsvirtualization

mapping

Privateregion

metadata

Application and/or Database File System

DMP, Drivers, Storage

As Figure 3-7 suggests, VxVM has two primary functional componepath and a metadata management function. In the I/O path, VxVM tI/O requests to virtual devices made by applications (usually file sys

read and write commands to the disks it controls, and issues the comoperating system device drivers, or to DMP if it is installed.

TheVxVM I/Opath uses parameter-drivenmapping algorithmsto traI/O requests into physical commands. The parameters specify whichof which disks represent which block ranges of which virtual devices.for a mirrored striped volume, the principal parameters are:

The disks containing the VxVM subdisks that make up the stripe

their relative positions in the plexes Starting disk block numbers and sizes for each subdisk

The stripe depth, or number of consecutive blocks that map to edisk

A closer look at the Storage Foundation core: the VxVNew in the Storage Foundation: multi-host coordinated sto


55/167

function also responds to external events that change the stora

such as disk failure, commands issued by administrators, and AVxVM-aware applications like the VxFS file system. Whatever tevent, t