Download - Storage Enhancements in Windows 2012 R2
Storage
Michael RüefliCloud ArchitectINSERTO AG, SwitzerlandEmail: [email protected]: @drmiruBlog: www.miru.ch
2012 R2
Addressing common Block Storage Issues•Costs (CapEx) per I/O vs. per GB
•Complexity (OpEx)
•Performance
•Scalability
Short Explanation on IOPS
IOPS (Input/Output Operations Per Second, pronounced eye-ops) is a common performance measurement used to benchmark computer storage devices like hard disk drives (HDD), solid state drives (SSD), and storage area networks (SAN).
SEQUENTIAL Access
RANDOM Access
How Server 2012 R2 can help•Storage Spaces
•Storage Tiering
•Scaleout File Server
•SMB 3.0
•Online Deduplication
•Encryption
•Big Data (ReFS)
•Ease of Management
•Automation
•VHDX Sharing
Storage Spaces
• Virtualization of storage with Storage Pools and Storage Spaces
• Storage resilience and availability with commodity hardware
• Resiliency and data redundancy throughn-way mirroring (clustered or unclustered) or parity mode (unclustered)
• Utilization optimized through thin and trim provisioning and enclosure awareness
• Integration with other Windows Server 2012 capabilities
• Serial Attached SCSI (SAS) and Serial AT Attachment (SATA) interconnects
Storage Spaces
WindowsVirtualizedStorage
Windows Application Server or File Server
Physical or virtualized deployments
PhysicalStorage
(Shared) SAS or SATA
Integrated with otherWindows Server 2012 capabilities
Storage PoolStorage Pool
File Server Administration
Console
Hyper-V
Cluster Shared Volume
Failover Clustering
SMB Multichannel
NFS Windows Storage Mgmt.
NTFS SMB Direct
Storage Space Storage Space Storage Space
ENTERPRISE-CLASS FEATURES ON LESS EXPENSIVE HARDWARE
Storage Spaces - Capabilities Overview Pooling of disks
Flexible resilient storage spaces
Native data striping maximizes performance
Enclosure awareness with certified hardware
Parallelized rebuild utilizes spare pool capacity
Physical Disks from
Shared SAS JBODs
Storage Pool
Storage Spaces
Data Copy 1
Mirror Space
Mirror Space
Parity Space
Mirror Space
Data Copy 2
Storage Spaces – Resiliency Options
2-Way MIRROR SPACE
STORAGE POOL
SIMPLE SPACE
Simple Space
2-Way MirrorSpace
3-Way Mirror Space
Parity Space Dual Parity Space
Number of data copies
1 2 3 2 + Parity 3 + Parity
Number of columns(default)
# Disks # Disks / 2 # Disks / 2 # Disks # Disks
Interleafe(default)
256KB 256KB 256KB 256KB 256KB
PARITY SPACE
Storage Space
Reduced Mean Time To Recovery
Mirror Spaces Rebuild Parallelized Recovery
Increased Throughput Optimized Disk Utilization Utilization of spare capacity
Hot Spare no longer necessary in R2
Performance improvements to
radically reduce MTTR
Rebuild Metric Measurement
Data Rebuilt 2,400 GB
Time Taken 49 min
Rebuild Throughput
> 800 MB/s3TB HDDs, 2-way, 4-column Mirror SpaceSource: Internal Testing, No Foreground Activity
DemoStorage Spaces
File System
Platform Storage – Local File System
ReFS File System Greater
compatibility with SATA standards
Resilience against power outage corruption
Online corruption recovery
Implements on-disk structures, such as the Master File Table, to represent files and directories.
New Storage Capabilities with Resilient File System (ReFS)
Scale to PetaByte+
File system enhancements
• Rapid recovery from file system corruption without affecting availability
• Resilient against power outage corruption
• Periodic checksum validation of file system metadata
• Improved data integrity protection
• Ideal for file server volumes
• Compatible with CSV
• If used as Hyper-V VM Storage, file stream integrity has to be disabled
NTFS improvements
• Rapid recovery from file system corruption without affecting availability
• Data corruption virtually eliminated through allocate-on-write
• Period checksum validation of file system meta-data
• Seamless data integrity protection
Resilient File System (ReFS)
CHKDSK
• Seconds to fix corrupted data
• No offline time when used with CSV
• Disk scanning process separated from repair process
• Online scanning with volume and offline repair
CONTINUOUS APPLICATION AVAILABILITY
100 Million Files 200 Million Files 300 Million Files0
50
100
150
200
250
300
350
400
Windows Server 2008 R2
Source: “Microsoft Internal Testing"
Storage Performance
The Performance Dilemma
90% of costs for 10% of critical workloads (peaks)
Storage tiering
Storage Tiering
Optimized Data Placement Pool consists of both SSDs and HDDs Tiers within a Storage Space Hot Data (SSDs), Cold Data (HDDs) Sub-File-Level Data Movement Complements write-back caching
Administrative Controlled Pinning Pin hot files to faster SSD tier Example: VDI Pooled VM’s VHD
Tiered Storage Spaces provides
persistent performance
improvement for hot data
The best of both worlds:
SSD Performance & HDD Capacity
Storage Space
HDD Tier
Cold Data
SSD Tier
Hot Data
400GB EMLC SAS SSD
4TB 7200RPM SAS HDD
Virtual Machines
Write-Back Cache
Random Disk Access Inherent to storage access patterns with virtualized deployments SSDs provide higher IOPs for random access
Improved performance for real-world workloads SSDs absorb random writes at high IOPS levels Seamless integration and familiar management
Windows Server 2012 R2 maximizes performance and
capacity
The Spaces Write-Back Cache absorbs spikes in random write
activity
Virtual Machines
Storage Space
HDD Tier
Cold Data
SSD Tier & WBC
Hot Data
400GB EMLC SAS SSD
4TB 7200RPM SAS HDD
Write-Back Caching
Write-Back Caching (WBC)Complements Tiering
Tiering – Balances capacity & performance WBC – Short-term absorption of I/O bursts
Seamless Integration & Management
Utilizes available SSD capacity from pool WBC enabled by default Configurable through PowerShell
Hyper-V Compute Nodes
Storage Space
HDD Tier
Cold Data
SSD Tier & WBC
Hot Data
400GB EMLC SAS SSD
4TB 7200RPM SAS HDD
Reads/Writes Accumulates Data Activity
Seamless
Integration
DemoStorage tiering
Storage Deployment Approaches
Balanced
• Mix of HDD & SSD
• High Capacity HDDs
• High Performance SSDs
• Best IOPS/$
How to build a deployment with:• 18 TB Capacity• 200’000 IOPS (8k,
random)
HDD Only Deployment• Meet Capacity with 9
HDDs• Miss IOPS by 198’200
SSD Only Deployment• Miss Capacity by 16.4
TB• Meet IOPS with 4 SSDs
Balanced Deployment:• Meet Capacity with 8
HDDs• Meet IOPS with 4 SSDs
OptionCapacit
yIOPS Cost
SSD Only (45)
16 TB ~1’000’000
$63,000
Mixed(8 + 4)
19.6 TB ~200’000
$8’000?
3TB SAS HDDs, 150 IOPS, $350 MSRP
800GB SAS SSDs, 50,000 IOPS, $3,000 MSRP
Demo Hardware Configuration
Industry Standard HardwareHyper-V Hosts 4
Storage Nodes 2
Hyper-V RAM 1,2 TB
Storage Capacity (SSD) 700 GB (dual mirroring)
Storage Capacity (SAS) 7 TB (dual mirroring
Storage Access Protocol SMB3 Multi-Channeling (2 x 10Gbps)
Max. write IOPS (8K, 100 % random) / Max. Troughput
180’000 / 1022 MB/s
Max. read IOPS (8K, 100% random) / Max. Troughput
220’000 / 1100 MB/s
Total CapEx $ 97’000.-
Demo Hardware Layout
>200’000 IOPS>2GB/s Throughput<100K CHF
DemoStorage Performance(with Hyper-V VM on SMB3.0 Share)
Component Overview
Industry standard commodity storage hardware Shared SAS JBOD arrays with power and data-path redundancy Dual-Port SAS drives
Sampling of available Spaces Certified Hardware
See www.windowsservercatalog.com under the “Storage Spaces” Category
Some Advice (from the field)….•Only use Enterprise SSDs for productive Workloads
•Check and Respect HCL (end-to-end)
•Calculate your required Tier Sizes
•Leave Spare Space on Pools for Rebuilds
Data Deduplicati
on
Efficient storage through Data Deduplication
VHD Library
Software Deployment
ShareGeneral File
Share
User Home Folder (My Docs)
0% 20% 40% 60% 80% 100%Average savings with Data Deduplication by workload
type
ENTERPRISE-CLASS FEATURES ON LESS
EXPENSIVE HARDWARE
Maximize capacity by removing duplicate data
• 2:1 with file shares, 20:1 with virtual storage
• Less data to back up, archive, and migrate
Increased scale and performance
• Low CPU and memory impact • Configurable compression schedule• Transparent to primary server
workload
Improved reliability and integrity
• Redundant metadata and critical data
• Checksums and integrity checks• Increase availability through
redundancy
Faster file download times with BranchCache
Source: “Microsoft Internal Testing"
• Data Access Zero impact on new written files No impact on access to user documents (e.g. office) 1-1.5x time to copy 1 large file (e.g. VHDs) <1x time to copy multiple large files (e.g. VHDs) with high duplication
• Deduplication Data optimized at ~100GB/hour for a single 2TB volume/disk using a single
CPU core and <1GB of RAM Scales out: Multiple volumes can be processed in parallel with additional CPU,
memory and disk resources
Performance and Scalability
DemoEvaluating Deduplication
Evaluating Deduplication
Scaleout Fileserver
Windows Scaleout File Server Cluster
• Continuously Available Storage Platform Supports VM migrations Storage moves with VMs Resiliency to disk, link, controller,
power, JBOD, and node failure
• Cluster Shared Volume Unified namespace scales linearly
as deployment grows
Hyper-V Compute Nodes
Mirror
7 6 8 G b / s S h a r e d S A S L i n k s
60-baySAS Array
60-baySAS Array
60-baySAS Array
60-baySAS Array
MirrorMirror Dual Parity
SMB
\\SRV\VDI_Mktg
\\SRV\Ops
\\SRV\DB
SMB
… …
Scaling the Windows File Server Cluster
Clustered Storage Spaces
Physical or Virtualized Workloads Hyper-V Compute Nodes
High Speed Network(10GbE/InfiniBand)
Unified Namespace (CSV)
60-bay Shared SAS JBOD Arrays
Clustered File Servers with 10GbE/InfiniBand
768Gbps Shared SAS Links
768Gbps Shared SAS Links
768Gbps Shared SAS Links
0.96PB
0.96PB
0.96PB
768Gbps Shared SAS Links
0.96PB
\\SRV2\VDI_Mktg \\SRV3\Ops\\SRV1\DB
SMB SMBSMB
Mirror
Mirror
Mirror
Mirror
Mirror
Mirror
Mirror
Mirror
… … … …
\\SRV4\Infra
Server Message Block
(SMB) 3.0
EMC's view on SMB3
http://www.emc.com/collateral/hardware/solution-overview/h11912-vnx-with-ms-smb-3.0-so.pdf
• Top-to-bottom performance analysis and tuning SMB server optimization New tracing instrumentation to
identify bottlenecks Substantial refactoring of SBM2 client
as well
• Windows Server 2012 at 98% of DAS transactional perf Identical servers and storage 1Gbps Ethernet v. 4Gbps FC 28 x 10KRPM HDD
SMB SMB + DAS0
500
1000
1500
2000
2500
3000
3500
4000
4500
1175 (27%)
4270 (98%) 4315 (100%)
SQL TPC-C transactions/second
SMB 3.0 Direct-attached storage (DAS) Equivalence
Network fault tolerance with SMB Multichannel
NetworkSMBclient
SMBserver
Filecopy
NIC NIC
NIC NIC
Filecopy
• Automatic detection and use of multiple network connections between SMB client and server
• Helps server applications be resilient to network failure
• Transparent Failover with recovery of network failure if another connection is unavailable
• Improved throughput
• Bandwidth aggregation through NIC Teaming
• Multiple nodes/CPUs for network processing with RSS-capable network adapters
• Automatic configuration with very little administrative overhead
CONTINUOUS APPLICATION AVAILABILITY
File Client
SMB ClientSMBBuffer
File Server
Application
NIC Driver
Transport Protocol Driver
With RDMA
Improved network performance through SMB Direct (RDMA*)
Without RDMA
Application
NIC Driver
SMB ServerSMB Client
Transport Protocol Driver
NIC Driver
TransportProtocol Driver
AppBuffer
SMBBuffer
OSBuffer
Driver
Buffer
SMBBuffer
OSBuffer
Driver
Buffer
SMB Server
NIC Driver
TransportProtocol Driver
AppBuffer
SMBBuffer
rNIC* rNIC* NIC AdapterBuffer NICAdapter
BufferAdapterBuffer
AdapterBuffer
iWARP
InfiniBand
ENTERPRISE-CLASS FEATURES ON LESS
EXPENSIVE HARDWARE
• Higher performance through offloading of network I/O processing onto network adapter
• High throughput with low latency and ability to take advantage of high-speed networks (such as InfiniBand and iWARP)
• Remote storage at the speed of direct storage
• Transfer rate of around 50 Gbs on a single NIC port
• Compatible with SMB Multichannel for load balancing and failover *Remote Direct Memory Access *e.g. Intel® NetEffect™ Server Cluster Adapter Family
• Highly available, shared data store for SQL Server databases and Hyper-V workloads
• Increased flexibility, and easier provisioning and management
• Ability to take advantage of existing network infrastructure
• No application downtime for planned maintenance or unplanned failures with failover clustering
• Highly available scale-out file server
• Built-in encryption support
File Server Cluster
Application storage support – SMB 3.0
Cluster Shared Volumes
Single File System Namespace
SMBSingle Logical Server \\Foo\
Share
SAN
WindowsVirtualizedStorage
PhysicalStorage
Storage PoolStorage Pool
Storage Space Storage Space Storage Space
ENTERPRISE-CLASS FEATURES ON LESS
EXPENSIVE HARDWARE
Hyper-V ClusterMicrosoft SQL Server
SMB Transparent Failover
Failover share – connections and handles lost,temporary stall of I/O
2
Normal operation1
Connections and handles auto-recoveredApplication I/O continues with no errors
3
Win
dow
s Serv
er
file
serv
er
clust
er \\foo1\share1 \\foo2\share1
\\foo\share
SQL Server
• High-performance, continually available fileshares for business critical applications
• Failover transparent to server applications with zero downtime and with only a small I/O delay
• Support for planned moves, load balancing, operating system restart, unplanned failures, and client redirection (scale-out only)
• Resilient for file and directory operations
• All servers involved should have Windows Server 2012
CONTINUOUS APPLICATION AVAILABILITY
SMB Direct Performance Enhancements(R2)
Scale-out File Server
SMB Direct
Hyper-V
VHD50% improvement for small IO workloads
Efficiency
Increased efficiency and density of hosting workloads with small I/O’s such as OLTP database in a VM
Optimizes 40Gbps Ethernet and 56Gbps InfiniBand
Performance
50% improvement for small IO workloads with SMB over RDMA
Increased 8KB IOPs from ~300K IOPS to ~450K IOPS per interface
SMB
Scale-out File Server
Hyper-V host
Optimized Rebalancing of Scale-out File Server Scale-Out File Server clients are now redirected to the “best” node for access
Avoids unnecessary redirection traffic Driven by ownership of Cluster Shared Volumes SMB connections managed per share
No longer per file server Dynamically moves as CSV volume ownership
changes Clustering is now also balancing CSV volumes
automatically Automatic behavior, no administrator action
File Server 2File Server 1
Storage Spaces
\\SOFS\Share2
Share1 Share2 Share1 Share2
CSV and SMB shares automatically rebalanced
Improved network efficiency through
drastically reducing redirection traffic
DemoScale Out File Server & SMB 3.0
Offloaded Data
Transfer (ODX)
Offloaded Data Transfer (ODX)
OffloadCopy
Request Token
Write Request
TokenSuccessful Write Result
External Intelligent Storage Array
Virtual Disk Virtual Disk
Actual Data
Token
IMPROVED PERFORMANC
E,MORE CHOICE
Benefits• Rapid virtual machine provisioning
and migration• Faster transfers on large files• Minimized latency• Maximized array throughput• Less CPU and network use• Performance not limited by network
throughput or server use• Improved datacenter capacity and
scale
Offloaded Data Transfer (ODX)Token-based data transfer between intelligent storage arrays
iSCSI Target Server
File and Block Access
iSCSI network boot Typically used for
Hyper-V host boot and HPC boot
Rapid deployment Quick recovery Reduce capital
expenditures and operating expenditures
Ease of management
iSCSI Target Server with Network Boot
iSCSI Target Server
VHDX Support
• Provision LARGER (up to 64TB) LUs• Online expand/shrink of LUs• Provision dynamically-growing LUs
SMI-S Support
• Fully manage iSCSI Target Server using SMI-S
• Empowers full end-to-end management with System Center Virtual Machine Manager
• Standards-based management
iSCSI Target Server
SCVMMSM
I-S
iSCSI Target Server is highly scalable and flexible using
VHDX
Manage iSCSI Target Server out-
of-the-box with SCVMM
iSCSI Target Server delivers a
comprehensive management experience
NFS High availability storage for UNIX, Linux,
VMware
High availability storage for UNIX, Linux, VMware
Transparent failover support for NFS v3 or 4.1 clients
UNIX, Linux, and VMware support
New RPC-XDR runtime infrastructure and DLLs make NFS failover faster than ever
DemoNFS – Transparent Failover
DemoStorage Management
Native & via SCVMM
Encrypted cluster
volumes
Encrypted Cluster Shared Volumes (CSV)
• BitLocker encrypted cluster disks
Support for traditional failover disks
Support for Cluster Shared Volumes
• Cluster Name Object (CNO) identity used to lock and unlock Clustered volumes
• Enables physical security for deployments outside of secure datacenters
Branch office deployments
• Volume level encryption for compliance requirements
• Negligible (<1%) performance impact
• Application-consistent shadow copies for server application data stored on Windows Server 2012 file shares
• Backup and restore scenarios
• Full integration with VSS infrastructure
• Data Backup
VSS for SMB File Shares
Volume Shadow Copy Service
\\fs\fooData volume
\\fs\foo@t1Shadow
Copy
Backup Server
Application Server
File Server
File Share Shadow Copy Agent
Coordinate Shadow Copy
Create Shadow Copy
Create Shadow Copy
Request Shadow Copy
VSS Providers
BackupA
B
C
D
E
Read fromShadow CopyShare
G
Relay Shadow Copy
request
Backup Agent
Volume Shadow Copy Service
File Share Shadow Copy Provider
F
VHDX Sharing
Guest Clustering
Guest Clustering – via VHDX Sharing
•Dramatically simplifies VM guest clustering
•No vSAN or passthru disks required
•Attach a highly available VHDX to multiple VMs
• Supported on CSV and SMB 3.0 Scaleout Share
• Supported on standalone and clustered Hyper-V hosts
Hyper-V Node Hyper-V Node
DemoGuest Clustering via
shared VHDX
Windows Storage
Server 2012 R2
Windows Storage Server 2012 R2
• Windows Server for NAS appliances that are preconfigured and enterprise ready
• Ideal file-based storage solution that is fault tolerant, continuously available, scalable, and cost effective
• Workgroup and Standard editions
• Multiple vendors
Workgroup Standard
NAS Segments Entry level50 users
Scale upContinuous availability
Architecture/maximum memory X64 – 32 GB X64 – 2 TB
Active Directory Lightweight Directory Services
Yes Yes
CPU sockets/virtual machines 1 + 0 2 + 2
Hardware RAID Yes Yes
Disks 6, No external SAS Unlimited
iSCSI Software Target Yes Yes
File sharing (SMB/NFS) Yes Yes
Microsoft BitLocker Yes Yes
Data Deduplication No Yes
Hyper-V No Yes
BranchCache – Hosted Cache No Yes
Clustering No Yes
DHCP, DNS, and WINS No Yes
Active Directory DC, Certificate, Federation, Rights Management, Application Server, Network Policy, Terminal Services, WDS, Fax Server
No No
IMPROVED PERFORMANC
E,MORE CHOICE
Growing ecosystem of hardware partners
Windows Cluster in a Box ODX SMB Direct
Quanta Computer
HP
Wiwynn
Violin Memory
Dell EqualLogic
IBM
HP
NetApp
Fujitsu
Intel
LSI
XyratexXio
EMC
Hitachi
Fujitsu
NEC
Mellanox
Intel
Chelsio
HA-DAS.com
Windows Storage Server 2012
DELL
XTORE
MaXXan
AREA
Western Digital
LACIE
Unibrain
CM
Compellent
Pyramid
Thomas-Krenn.AG
Broadberry
Pivot 3
N-TEC
AMAX
ABERDEEN
BOSCH
Acma
DNF
Storageflex
BUFFALO
Variel
WINCHESTER SYSTEM
FujitsuSIEMENS
IBM
HP
NEC
IMPROVED PERFORMANC
E,MORE CHOICE
Windows File Server Cluster Solution PartnersCertified Storage Spaces HardwareHigh-Performance Servers and Storage
http://www.raidinc.com/ http://www.dataonstorage.com/http://www.fujitsu.com/fts/
http://www.quantaqct.com/
http://www.supermicro.com/
StorSimple
StorSimple's Cloud-integrated Storage (CiS)
StorSimple's Cloud-integrated Storage (CiS) has been helping customers do this for over a year now and proven it's mettle working in the data centers of large enterprises. CiS on-premises SAN storage exports iSCSI LUNs to servers and connects on the back, or device, side to enterprise cloud storage for storing snapshot, backup, archive and unstructured, dormant data. CiS is a hybrid SAN array having flash SSDs and hard disk layers but it adds a third slower, higher-latency enterprise cloud storage storage layer. It’s 3-tier design is perfectly matched for managing the masses of unstructured data that IT workers wrestle with.
StorSimple Solution Operation
SSD Deduplicated
SASDeduplicatedCompressed
CloudDeduplicatedCompressedEncrypted
SSDLinear TierA B C A B D E
C D E
D E
E
StorSimple Appliance ModelsSca
labili
ty /
Perf
orm
an
ce
Capacity
552010-50TB* usable local300TB max capacity
752020-100TB* usable local500TB max capacity
* Denotes usable local storage capacity with compression and de-duplication, varies by use case
* Additional details about appliance specifications can be found at: http://storsimple.xyratex.com/storsimple/specifications
50202-10TB* usable local100TB max capacity
70204-20TB* usable local200TB max capacity
Q&A
© 2013 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries.The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.
Thank You!
Michael RüefliCloud ArchitectINSERTO AG, SwitzerlandEmail: [email protected]: @drmiruBlog: www.miru.ch
Walter PitrofSr. Technology Solution ProfessionalMicrosoft SwitzerlandEmail: [email protected]