disk technologies in high-end storage arrays
TRANSCRIPT
Presented at the THIC Meeting at the Sony Auditorium, 3300 Zanker Rd, San Jose CA 95134-1940
April 19-20,2005
Disk Technologies in High-End Storage Arrays
Roberto BasilioHitachi Data Systems
750 Central Expressway, Santa Clara, CA, 95050-2627Phone: +1-408-970-5990 FAX: +1-408-562-5477
E-mail: [email protected]
Topics
• Introduction • Hard Disk Drives
– Evolution and directions• Storage Systems
– Architectures and Deployment • What’s next
– New technologies and deployment in large enterprise
Hitachi, Ltd. Worldwide Storage Group
Two Powerful Storage Businesses:• HDD Business/ Hitachi Global Storage Technologies
• Storage Systems & Solutions BusinessHitachi,Ltd. & Hitachi Data Systems
Leveraging Storage Technology Leadership:
Expanding and Investing in Emerging Markets:
• Enterprise Storage Platforms, Midrange Storage Systems, HDDs, Storage Management Software,Consulting Services
• Investing over $500M annually• Compliance Solutions, Rich Media,Mobile Music, Game Players, “Terabyte Home”
Disk Evolution
Disk Recording Densities Continue to Grow
Server HDD Transition – RPM History
4,923 3,809 1,894
4,287 7,546 11,393 13,003 11,796
5,0112,472 1,711 0
51,743
6,88815,195
15,276 16,21716,265
1,288 2,368 4,628
16233
268
220
0
5,000
10,000
15,000
20,000
25,000
1995 1996 1997 1998 1999 2000 2001 2002 2003
<= 5400 7200 10K 15K
RPM Transitions typically quick after new Dominator• 7200 RPM eliminated 5400 RPM 2 years after it reached >50%
• 7200 RPM was very quick to dominate the market after only 2+ years• 10K RPM domination in 2000 led to 7200 RPM extinction 2 years later
• 10K RPM took approx. 3+ years to become dominate• 15K RPM has now been in the market 5 years.
• Adoption slowed during the 2001 economic downturn• Many customers beginning to position 15K to become dominate• If history holds: 10K also destined to fall off relatively quickly by ’07 or ‘08
Historical data from IDCTraditional 3.5” Enterprise HDD, WW Unit Shipments
Strong Growth Ahead for Consumer Electronics
Information Technology
M Units/Year
SFF
1.01.82.5
3.5
Consumer Electronics
0
50
100
150
200
250
300
350
400
450
'02 '03 '04 '05 '06 '07 '08Source : Hitachi GST estimates
3.5” CE
Notebook PC
DesktopIT
CE
Traditional Server
Desktop PC
Notebook PC
PDA, Portable AudioPMC Car Navigation, iVDRCar Audio, DVR HDD Recorder, Gaming
Enterprise
Enterprise ATA
2.5
1.8
19%22%
47%
50%
56%
34%
0
2000
4000
6000
8000
10000
12000
14000
2004 2005 2006 2007 2008 2009
23128 24800 25800 26800 27800 28800
28114241
61179586
1164714500
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
50000
2004 2005 2006 2007 2008 2009
Enterprise by Application and Interface (Hitachi GST – Feb ’05)
CAGRs: 2004-2009 UNIT TB’sAll Enterprise: 11% 41%Enterprise ATA: 39% 76%Traditional Serve 5% 25%
17951 1657310100
4575862 328
848616109 20596 21984
4986 6328 7214 6116 6343 64883393
54458820
1106414500
1293
0
5000
10000
15000
20000
25000
30000
35000
40000
45000
50000
2004 2005 2006 2007 2008 2009
All Units by Interface
CAGRs: 2004-2009IDE / SATA: 39%FC / LCFC: 5%SAS: 84%SCSI: (55%)
Quick SAS Transition anticipatedContinued strong SATA GrowthAll Units by Application
All TB by Application
All Server* by RPM and Form Factor (Hitachi GST – Feb’ 2005)
CAGRs: 2004-2009SFF, 15K RPMSFF, 10K RPM: 104%3.5”, 15K RPM: 23%3.5”, 10K RPM: (33%)
17259 1531410550
2412
5793 8246
11896
17688
17236 16416
3320 656610036 10526
528 1858
0
5000
10000
15000
20000
25000
30000
2004 2005 2006 2007 2008 2009
3.5” 15K transition accelerating SFF Ramp slower than original projections
23052 23560 2244620100
17236 16416
3354 6700 10564 12384
0
5000
0000
5000
0000
5000
0000
2004 2005 2006 2007 2008 2009
1
1
2
2
3
CAGRs: 2004-2009
SFF: 115%3.5”: (7)%
*Server: consists of SCSI, FCAL, and SAS
Server* Units by RPM Server* Units by Form Factor
Storage Growth Drivers
Technology Trends and Drivers• Traditional large SMP architectures to diminish in numbers -
Server technology is moving from enterprise to clusters to blades– What does this mean to storage?
• Higher bandwidth requirements• High data accessibility (equivalent to high port counts today)• Clustered file system technology• Cluster partitioning
• New computing model requires many more server images to access a common storage pool
– Increasing need for connectivity and safe multi-tenancy on storage– Increasing need for scalable performance and QoS – Data Management workload (snapshots, replication, backup, HSM) need to
be offloaded from the servers • Fibre Channel, S-ATA, SAS, ?
– Common electrical interface will influence the decision• ATA disks will become the new high-end HDD
– Serial ATA is the growth tier drivers include Virtual Tape & Disk to Disk– Biggest “Pain Point” manual movement of data between Tiers– something else will fulfil the role of the low-end high volume disk– A shift in form factor - smaller
Storage Growth Drivers• Disaster Recovery
– Creating Distance Needs– High Availability
• Regulations– Sarbanes-Oxley Act compliance alone expected to reach US$6.1 billion in 2005,
according to AMR Research – Basel II requires increase in capital reserves for operational risk
• Consolidation– Especially with Servers
• Email & Desktop– Video & audio email (multi- media), Desktop Back up
• Internal Requirements– What goes to SAN/NAS/DAS
• Getting closer to the data– Data Warehousing “Get close to the customer”
• Application & Database Growth– Oracle, SAP – the usual suspects
• iSCSI & FC over IP –– gets SAN ’s out of the data center and opens up more ways for the use to increase
• But still low in the current users priorities
Storage Subsystems Technology
Distributed Architecture Centralized Architecture
• Higher functionality, centralized design
• Service multiple applications, many users
• Multiple processor controllers, shared global cache
• Shared Bus, Massively Parallel Crossbar Switch or Matrix contraption-based implementation
• Lower cost, granular design
• Service a specific application, few users
• Dual processor controllers, dual caches
• Bus-based implementation
Storage Architectures – Distributed vs. Centralized
Centralized Architecture
Fibre Channel Network
ESCON/FICON Network
IP LAN/WAN Network
SAN SAN SANSAN SAN SAN
S/390
S/390
S/390
NAS NAS iSCSl
S/390
NAS
Extensive, Multi-Protocol Connectivity
Future in High-end
• What comes next and what is going to be used in the next five years?– Virtualization and management– New connectivity
A look at what is hot today• Tiers, Tiers, Tiers….for both SAN & NAS
– Serial ATA is the growth tier drivers include Virtual Tape & Disk to Disk– Biggest “Pain Point” manual movement of data between Tiers– Missing Tier: IP SANs > sluggish adoption within the Fortune 1000
• ILM Still “Interesting” ; Implementations are Sparse– Potential to automate data movement of unstructured data is viewed as entry
• Virtualization: A “Momentum Play”– HDS, IBM & smaller firms
• NAS is a starting point for Virtualization, ILM
• Storage Management Software “Challenged”– Virtualization adding value for core Management needs– Unbundling of the purchase causes some surprises in budget allocations – Very “Tactical” functionality being implemented ie Reporting
• Too Early: WAFS, Storage Grids, Thin Provisioning, Fabric Intelligence
• OS Functionality more robust? – Volume Management, Multi-pathing
Enter Storage Virtualization
“An act of integrating one or more (back end) services or functions with additional (front end) functionality for the purpose of providing useful abstractions. Typically virtualization hides some of the back end complexity, or adds or integrates new functionality with existing back end services.”
The SNIA Dictionaryhttp://www.snia.org/education/dictionary/v/
defines virtualization as
The Innovator’s Dilemma Clayton M. Christensen
• Innovators who continually enhance their products are ambushed by disruptive technologies that are “good enough” – SATA storage is a disruptive technology
• Chasing disruptive technologies will compromise innovation– Requires a move to another dimension
• Virtualization changes the dimensions of the storage market by separating the storage controller from the disks– The real value is in the controller– Disks are becoming a commodity– Unbundling the storage controller enables more choices
Storage Controller
Storage Controller
What is the promise of Storage Virtualization?
• View the entire SAN as one pool of storage independent of physical location
• Mask the differences between heterogeneous devices• Simplify control and management under a single interface.• Centralize storage volume management on the SAN • Dynamically allocate capacity to the applications that need it • Facilitate matching storage resources to business requirements
Requirements For A Comprehensive Storage Virtualization Solution
• Volume Mapping and data mobility across heterogeneous vendor platforms
• Support multiple platforms with standard protocols– Direct attach, Fabric, FC, ESCON, FICON, NAS
• Does not introduce another layer of complexity• High performance and low latency• Scalable to thousands of host connections and petabytes of storage • Secure multi-tenancy for thousands of users sharing the same
physical assets• Logical partitioning for Quality of Service• Common end to end management for heterogeneous vendor
platforms
Virtualization
Where Should Virtualization Reside?
SAN
MainframeHostHostHost Virtualization in the host?VirtualizationVirtualization Virtualization DFSMS
In the SAN?Virtualization
VirtualizationVirtualizationStorage Controller Storage Controller
In the Device?Virtualization
Challenges of Inband SAN Virtualization
SAN
MainframeHostHost
VirtualizationVirtualizationStorage Controller Storage Controller
Host
Virtualization
Intercept I/O
Storage Controller Functions
• Additional complexity• SAN only attachment• Increased latency• Limited data mobility
functions• Lacks scalable cache and
access ports• Requires APIs• Violates end-to-end
security
• Agents in the host intercept I/O
• Storage Router redirects the I/O
• State is kept in the Switch and creates a vendor lock in
• Re-invent storage controller in the switch
• Lacks scalable cache and access ports
• Requires APIs and/or standards
• Violates end to end security
SAN
HostHost
VirtualizationVirtualizationStorage
ControllerStorage
Controller
Storage Route+Software
Storage Controller Functions
Agent Agent
Application Blade
Challenges of Storage Router Virtualization In the SAN
Virtualization
Virtualization
Virtualization
Virtualization in the Storage Controller
SAN
Mainframe
Storage Controller Storage Controller
HostHostHost
Virtualization
• No added complexity or latency between the host and storage
• Virtualization is done behind the Controller
• External storage attaches through standard Fibre Channel protocols
Connectivity – What is the future
Technology Trends & Drivers Fibre Channel – Roadmap (www.fibrechannel.org)
Product Naming
Throughput (MBps)
Line Rate (Gbaud)
T11 Spec Completed
(Year)
Market Avaiability
(Year)
1GFC 200 1.065 1996 1997
2GFC 400 2.125 2000 2001
4GFC 800 4.25 2003 2005
8GFC 1,600 8.5 2006 2008
16GFC 3200 17 2009 2011
32GFC 6400 34 2012 Market Demand
64GFC 12800 68 2016 Market Demand
128GFC 25600 136 2020 Market Demand
10GFC 2400 10.52 2003 2004
*Base2 used throughout all applications for Fibre Channel Infrastructure and devices. Each speed maintains backward compatibility at least two previous generations (I.e., 4GFC backward compatible to 2GFC and 1GFC)
**Base10 commonly used for ISL, core connections, and other high speed applications demanding maximum bandwidth
Base 2
Base 10
Technology Trends & Drivers – cont.
InfiniBand (www.infinibandta.org)
Infiniband Roadmap
10
40
80
120
0
20
40
60
80
100
120
140
2003 2004 2005 2006Year
Gig
abits
per
Sec
ond
Single Data Rate
Double Data Rate
Quad Data Rate
Technology Trends & Drivers – cont.Ethernet
Ethernet Bit Rates
0
20
40
60
80
100
120
1998 2000 2005 2010
Year
Gig
aBit
per S
econ
d
Line Rate
Trunk Rates
Conclusions
• Larger and faster units will greatly leverage forms of virtualization
• Multi-connectivity within the same system is a requirement– Use the best connection for the application to serve
• Hard Disk technology less important as long as it gets the job done– FC, SAS, SATA will have their common space for a while
• More Virtualization– Any form, multiple vendors– Ease the storage management
