© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Agenda
� IBM disk Portfolio
� DS8700 – Introduction– Technology Overview– Detailed features– New Features and improvements– Raw performance – Easy Tiering– Metro Global Mirror Function
� DS8800 description
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
IBM Disk Portfolio 1Q2011
N seriesFor clients requiring:� NAS storage� Simple two-site high availability
SVC, Easy Tier, ProtecTIER, Information Archive, Storwize AppliancesFor clients requiring optimization of storage costs:� Storage technologies that add function, performance, ease of use, or efficiency to new or existing storage
Storage Optimizers
DS3000For clients requiring:� SMBs or branch office locations; cost sensitive; start at as small as
100s GB, up to low TB capacity
Entry
Storwize V7000For clients requiring:� 10s of TB of rack-mounted
storage with sophisticated software functions
� Breakthrough ease of use and management
� Non-disruptive migration from, or virtualization of, existing disk
DS5000For clients requiring:� Good cost/performance,
general-purpose storage� Need to add capacity to
existing configuration� 10s of TB of capacity
Midrange
SONASFor clients requiring:� Massive I/O, backup, or
restores � Consolidation, or scale large
numbers of clusters
XIVFor clients requiring:� Breakthrough ease of use and
management� High utilization with automated
performance (“hot spot”) management
� Virtually unlimited snapshots� Advanced self-healing
architecture
DS8000For clients requiring:� Advanced disaster recovery
with 3-way mirroring and System z GDPS support
� Continuous availability, no downtime for upgrades
� Best-in-class response time for OLTP or transaction workloads, flash optimization
� Single system capacity scalable to the PB range
Enterprise
FileBlock
Optimized for ‘open systems’ Optimized for z/OS and IBM i
© 2011 IBM Corporation
IBM®
System Storage®
DS8700Introduction
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Binary Compatibility
2004-6 2007-9
POWER5 POWER6POWER PC / POWER
1999-4
130 nm 90 nm 65 nm
L2 Cache
4.7GHz
AdvancedSystem Features
Distributed Switch
180 nm 130 nm
1.9GHzCore
Distributed Switch
Shared L2
1.5+GHzCore Shared L2
2.2GHzCore
Distribued Switch
2.2GHzCore
1.5 GHzCore
Distributed Switch
Shared L2
1.5 GHzCore
DS8000
DS8000 Turbo
DS8000
DS8000 TurboESS (Shark)ESS (Shark)
The IBM POWER processor has been behind the success of IBM enterprise storage beginning with the Enterprise Storage Server in 1999
Benefit: a steady, 10-year lineage of RAS improvements!
IBM POWER and DS8000 evolution
DS8700DS8700
FUTURE
POWER xxx
DS8XXXDS8XXX
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Why Move to DS8700
� Performance– Up to over 150% performance boost with new IBM POWER6-based controllers – New, faster PCI Express (PCI-E) internal fabric enables much higher performance and
scalability– Almost 70% faster ASIC on the device adapters– Increased FlashCopy performance in every metric– Increased SSD performance on sequential reads
� Availability– Single model, scalable via concurrent upgrade of all components– Shorter service windows with faster concurrent microcode updates– Better than 99.999% availability
Introducing the DS8700 (1/2)
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Why Move to DS8700 (2/2)
� Investment Protection and Scalability– DS8700 architecture allows for future expansion of additional controllers and future 8Gb
Host Adapters– SSD optimization and M/GM Multiple Session announced
� Management– Simplified management and application-aware FlashCopy
� Security– Full Disk Encryption enhancements address PCI-DSS compliance
Introducing the New DS8700 (2/2)
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8700 Hardware Overview
� 2-Way (Model 941)– Two dual processor servers (POWER6)
•Up to 128GB Cache (32, 64 or 128 GB)– 8 to 64 2Gb/4Gb FC/FICON (4 ports adapters)– 16 to 128 Disks– Physical capacity from 2,2TB up to 128 TB
� 4-Way (Model 941)– Two four processor servers (POWER6)
•Up to 384GB Cache (32, 64, 128, 256 or 384 GB)– 8 to 128 2Gb/4Gb FC/FICON (4 ports adapters)– 16 to 1024 Disks (with 4 expansion frames 94E)– Physical capacity from 2,2TB up to 1024 TB
� All models :– Disks :
•300 / 450 / 600 GB FC 15K rpm (Encrypted or Non Encrypted)•1 & 2 TB SATA 7200 rpm•73 / 146 GB SSD (Solid-State-Drives) (8 or 16)
– RAID 5, 6 & 10 support
© 2011 IBM Corporation
IBM®
System Storage®
DS8700Technology Overview
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Change in I/O structure :
� New IO Towers are directly connected to the servers via point-to-point PCI-e cables.– IO Towers no longer share common ‘loops’, they connect directly to each server
via separate cables and link cards.
� New Device Adapter pair III (to connect disks to internal servers) – Processor upgrade (500 to 750Mhz) to increase I/O performance
Change in servers :� 4.7 Ghz P6 570 CECs� Cache 32GB to 384GB� 4Gb/s FC / FICON� 2Gbs disk array
P5
P5
P5
P5
Disks
Disks
DS8300RIO I/O attach New I/O
attach
DS8700PCIe cable I/O attach
P6
P6 P6
RIO RIO
PCIe cables
P6
SAN
Changing from DS8300 to DS8700
P6 server
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8700 – IO Fabric
� Key difference: The RIO loop in DS8700 will be isolated– Unlike DS8000, DS8700 IO towers will be on a separate point-to-point
connection– This isolates the server communication from IO fabric failures, power supply
problems, and repair procedures.
� DS8700 will use a PCI-e IO Fabric instead of PCI-x. PCI Express reduces the impact of hardware failures
© 2011 IBM Corporation
IBM®
System Storage®
DS8700Detailed Features
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8000 SARC Cache Algorithm
� SARC : Sequential Prefetching in Adaptive Replacement Cache
� Best caching algorithms in industry
� Simplified Adaptive Replacement Cache (SARC)� Self-Learning algorithms� Adaptively and dynamically learn what data should be
stored in Cache based upon the recent access and frequency needs of the Hosts
� Adaptive Replacement Cache� Most advanced and sophisticated algorithms to
determine what data in Cache is removed to accommodate newer data
� Pre-fetching� Predictive algorithm to anticipate data prior to a host
request and loads it into Cache
� Benefits� Leading performance� Been proven to improve Cache hit by up to 100% over
previous IBM caching algorithms and improve I/O response time by 25%
� More efficient use of Cache� Intelligent caching algorithm profiles Host access
patterns to determine what data is stored� Need less Cache than competitors
0
0.2
0.4
0.6
0.8
1
0 64 128 192 256
Cache Size (GB)
Cache H
it R
atio
z/OS
Open
Lower cache-to-backstore ratios
Benefits of adaptive replacement caching
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8000 Cache Efficiency elements
Example : Two random 4KB blocks
blk1
blk2
DS8000
4KB slots
blk2
blk1HDS high-end
16KB slots
--wasted--
blk1EMC DMX-3/4
64KB slotsblk2
-----------------------------wasted-----------------------------
The more random the I/Os and the smaller the blocks,
the more cache that is wasted due to large cache slot sizes
DS8000 Utilizes Less Resources to Manage Cache
System cache is a key factor in application I/O
performance.
It is also relatively expensive and relatively small
compared to total drive capacity.
Optimizing cache use via efficient system
management allows the system to “do more with
less”.
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8000 logical configuration and virtualisation
6+P+S 146GB 6+P+S 146GB
An array site is a collection of 8 drives of the same type spread across both internal fabrics(loops) on the DS8000
An array is formed when the drives in an array site are formed into a RAID group (RAID5/6/10) and potentially a spare is assigned
A rank is created when an array is formatted for either CKD or FB
Ranks are divided into extents which are either 1GB (FB) or 1113 cylinders (CKD)
An extent pool is a collection of ranks all of which are managed by the same DS8000 server (0/1)
Volumes are created from one or more extents from a single extent pool
A LSS/LCU is a collection of volumes and has relevance mostly for copy services and CKD
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Volume allocation algorithms
The rotate volumes allocation algorithm will attempt to keep a volume on a single rank and will rotate the rank for each allocated volume among the ranks in the extent pool
The rotate extents allocation algorithm will spread a volume over all available ranks in the extent pool and will rotate the rank for each extent among the ranks in the extent pool. The starting rank for each allocated volume will also be rotated in a similar manner
Note : There is also a historical allocation method called legacy which may be present on very old DS8000s and involved allocating the volume on the rank with the most available space
Rotate extents is also referred to as Storage Pool Striping and is the recommended method for multiple rank extent pools
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
0
50
100
150
200
250
300
350
400
450
Before AMP With AMP
Th
rou
gh
pu
t (M
B/s
ec
)
6+P Raid-5
7+P Raid-5
DS8000: Adaptive MultiDS8000: Adaptive Multi--stream stream PrefetchingPrefetching (AMP)(AMP)
� Invented by IBM Almaden Research Lab
� Decides what, when, and how much to prefetch
� Provably optimal
� Doubles Sequential Read Bandwidth for a Single DS8000 RAID-5 Array
Specific Cache Algorithms
DS8000: Intelligent Write Caching (IWC)DS8000: Intelligent Write Caching (IWC)
�Invented by IBM Almaden Research Lab – multiple IBM patents
�Improves performance through better write cache management and better destage order of writesUp to 2X throughput for random write workloads
�Typical database workload throughput may improve 15 - 20%
�DS8000 SMP cycle usage reduced up to 12% via better data structures
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DB Open Performance
70/30/50 Load
(single 8 disk arrays)
15 K RPM Disks
0
2
4
6
8
10
12
14
16
18
20
0 500 1000 1500 2000 2500 3000
Throughput (IO/sec)
Av
era
ge
Re
sp
on
se
Tim
e
(ms
)
RAID 6 – IWC
RAID 6 – no IWC
RAID 5 – IWC
RAID 5 – no IWC
RAID 10 – IWC
RAID 10 – no IWC
Intelligent Write Caching – Some Performance Results
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Value proposition:
Increasing DS8000’s superior resiliency to support today’s lower-cost, less-
reliable, large-capacity drives
� RAID 6 - offers protection from double disk failures by using a second independent distributed parity scheme (dual parity)
– Can tolerate loss of data in two strips of any stride without data loss• 2 Disk Failures• 1 Disk Failure + 1 Media Error• 2 Media Errors
� Critical for larger drives (SATA/FATA), which have longer rebuild times and thus longer periods of potential exposure should two drives start failing
RAID 6 Support
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
RAID-6 versus other RAIDs
RAID-6 may be used with Enterprise or Near-line ATA disks
662.5% *2RAID-6, 5+P+Q
673.5%2RAID-6, 6+P+Q
250%1 or 2RAID-10, 4+4
487.5%1RAID-5, 7+P
Performance Write penalty (Disk ops)
Space efficiencyReliability (#Erasures)
* includes Spare Overhead
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Leading solution : performance, and scalability for disk mirroring
Disk Mirroring solutions for ESS and DS
Metro Mirror (Synchronous PPRC)
Global Copy (Asynchronous PPRC= Extended Distance)
Global Mirror (Global Copy + automatic "goto sync" using Metro) Mirror
Metro Mirror/Global Copy ( Cascading PPRC)
Metro Mirror/Global Mirror (Cascading PPRC with automatic "goto sync")
Primary Site(PPRC primary)
Metro Mirror
Secondary Site
(PPRC secondary)
Global Copy
or
Global Mirror
FlashCopy
Global Mirror
Primary Site(PPRC primary)
Metro Mirror
Intermediate Site(PPRC
secondary/primary)
FlashCopy
Remote Site(with 'safety copy')
Global Copy
Primary Site(PPRC primary)
Intermediate Site(PPRC
secondary/primary)
FlashCopy
Remote Site(with 'safety copy')
Metro Mirror
DS8000 Disk Storage Mirroring
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
FlashCopy to normally provisioned volumes
Physical Space
FlashCopy FlashCopy
physical copy
physical copy
Classic FlashCopy
normally provisioned
normally provisioned
normally provisioned
normally provisioned
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
FlashCopy to Space Efficient volumes
Physical Space
FlashCopy FlashCopy
physical copy physical copy
Space Efficient FlashCopy
space efficient
space efficient
repository volume
normally provisioned
normally provisioned
normally provisioned
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Standard Logical Volume (LV)
physical copy
Standard LV
normally provisioned
physical copy
normally provisioned
All capacity for logical volume is allocated when volume is created
Unmodified tracks still consume real capacity
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
One repository volumeholds all tracks for many
Track Space Efficientvolumes
Track Space Efficient LV – (FC SE)
space efficient
vitual spacenormally
provisioned
physical space
Track Space Efficient
physical copy
virtual copy
Real capacity allocated when Repository is allocated
Repository capacity not consumed until writes to TSE LV
Each track (1st time updated) generates a real track allocated in SE repository
Logical capacity is consumed as it’s allocated
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Extent Space Efficient LV – (Thin provisionning)
One extent holds only tracks
for one Extent Space Efficient logical volume
An ESE LV can have multiple extents
No real capacity consumed until 1st write to source volume extent
Each new extent updated (data written) generates a real extent allocated to an ESE volume
Real capacity is consumed on each 1st
write to an extentspace
efficient
vitual spacenormally
provisioned
physical space
Extent Space Efficient
physical copy
virtual copy
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
FlashCopy to Metro Mirror Primary – Establish Pair
Local A
Local B
duplex
pendingfull
duplex
Local Storage Server
Remote Storage Server
Metro Mirror
FlashCopy
• FlashCopy target can also be a synchronous Metro Mirror primary volume
• After the FlashCopy command completes, the data associated with the FlashCopy will be physically copied from the Metro Mirror primary to the Metro Mirror secondary.
• During this operation, the Metro Mirror relationship will go into a “Duplex Pending” state
• HyperSwap cannot function properly during this unsynchronized state until the replication completes
Prior to Remote Mirror FlashCopy
Remote B
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Remote Pair FlashCopy Simplifies High Availability
Metro Mirror
Local Storage Server
Local A
Local B
Remote Storage Server
Remote A
Remote B
full duplex
full duplex
FlashCopy
Establish
FlashCopy
Now with Remote Mirror FlashCopy
• Now a FlashCopy command is established both locally and to the Remote FlashCopy source
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8700 Full Disk Encryption Enhancements
New enhancements address Payment Card Industry (PCI-DSS) compliance requirements with improved Secure Key management options
– DS8000 is the only enterprise disk platform with self-encrypting drives• Advanced key management is compatible with existing tape encryption solution• No performance impact compared to standard, non-encrypting drives• Transparent to existing applications and disaster recovery environments• Standards-based key management• IBM Global Human Resources has deployed DS8000 FDE
– New enhancements help clients comply with PCI-DSS requirements• Dual platform key server supports secure key requirements• Also new Encryption Deadlock Recovery Key
Advanced security for data at rest
IBM is rated in the top 5 of “Most Trusted Companies
in [Data] Privacy”*
IBM is rated in the top 5 of “Most Trusted Companies
in [Data] Privacy”*
*Ponemon Institute and TRUSTe Rank America’s Most Trusted Companies in Privacy, 9/16/09
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
z/OS Distributed Data Backup
System z
FDR CKD Backup app.
Multiple distributed
environments
z/OSmainframe
environment
•One single backup environmentfor mainframe data and distributed data•No Separate Backup serversNo TCP/IP Data Transfer•Removes backup server and open tape from infrastructure
MAINFRAME PRODUCTION, BACKUP AND DISTRIBUTED
BACKUP
1FDRSOS on Systems z takes
non-disruptive backup from copy of open system volumes & puts backup data to z-attached tape, disk or ProtecTIER VTL for dedup
open volumes now have two connections one for production read/write (from distributed servers) and the other for backup read/write (from z)
SCSI architecture / FCP
CKD Architecture / FICON
Mainframe and open
data
z/OSTape
TCP/IPLAN/WAN
AIX®
Windows
FBA Architecture / FICON
2FDRSOS FBA Backup app.
UPSTREAM z/OS Backup app Server
1
2
Multiple separate backup environments for
mainframe data and distributed data
Separate open systems backup servers
Data moving across TCP/IP links
Prior to z/OS Distributed Data Backup
System z
FDR CKD Backup app.
Multiple distributed
environments
z/OSmainframe
environment
PRODUCTION AND BACKUP
SCSI architecture / FCP / SAN
CKD Architecture / FICON
Mainframe data
z/OSTape
TCP/IPLAN/WAN
AIX
Microsoft®
Windows®
PRODUCTION AND BACKUP
Open dataOpenTape
Multiple distributed
backup servers
No Backup data on TCP/IP links
© 2011 IBM Corporation
IBM®
System Storage®
DS8700 & code R5.0New Features & Improvements
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011 GUI enhancements
Updated every 60sec.
Number of racks shown matches racks in the storage
unit
New System Summary Panel
– Overview tab displays system configuration
and real-time performance data for a 5 minute
sampling interval
New System Summary Panel
– Overview tab displays system configuration
and real-time performance data for a 5 minute
sampling interval
Improved and Simplified Management
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
� Hardware Explorer tab shows system hardware and a mapping between logical configuration objects and drives (e.g., drives belonging to an extent pool or array)
� Real images show exactly what bays are populated with what devices
� Hardware Explorer tab shows system hardware and a mapping between logical configuration objects and drives (e.g., drives belonging to an extent pool or array)
� Real images show exactly what bays are populated with what devices
R5.0R5.0
System Summary Panel – Overview Tab
GUI enhancements
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
� IBM Tivoli Storage FlashCopy Manager provides replication integration between enterprise applications and IBM disk systems
– Similar to NetApp Snapshot™ and NetApp SnapManager®?
FlashCopy�Create instant application copies for backup
or application testing
�Many replication options including
incremental (only changed blocks) or
space-efficient copies (“snapshots”)DS8000
FlashCopy Manager �Integrated, instant copy for critical
applications
�Virtually eliminate backup windows
�Rapidly create clones for application testing
�View inventory of application copies and
instantly restore
XIV DS3/4/5 SVC
Application-aware FlashCopyTivoli Storage FlashCopy Manager
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Code Upgrade
� Concurrent Code Load Improvements :• Allow part of current code load process to be completed prior to activation of
code without any loss of access or quiescing of any resource (no loss of redundancy).
• This reduces the activation phase of the code load process time required.
• Intended to provide a less than 2 hour activation phase of the main components of the DS8000.
© 2011 IBM Corporation
IBM®
System Storage®
DS8700 Raw performance
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8700 Performance Matrix (System z workloads)
DS8K Metric DS8700 DS8300 % Chg
FICON 4K Read Hit (IOs/sec) 364K 232K 56%
FICON 4K Write Hit (IOs/sec) 145K 120K 20%
zHPF 4K Read Hit (IOs/sec) 420K 344K 22%
zHPF 4K Write Hit (IOs/sec) 162K 124K 30%
FICON DB zOS (IO/sec) 181K 124K 45%
FICON Cache Hostile (IO/sec) 142K 89K 59%
FICON Cache Friendly (IO/sec) 206K 142K 45%
zHPF DB zOS (IO/sec) 201K 165K 21%
zHPF Cache Hostile (IO/sec) 156K 109K 43%
zHPF Cache Friendly (IO/sec) 227K 189K 20%
FICON 4K Read Miss (IOs/sec) 129K 64K 101%
FICON 4K Read-Write Miss (IOs/sec) 89K 62K 43%
FICON 4K Write Miss (IO/sec) 71K 49K 44%
zHPF 4K Read Miss (IOs/sec) 128K 86K 48%
zHPF 4K Read-Write Miss (IOs/sec) 90K 63K 42%
Sequential Read BW (GB/sec) 9.3 4.1 126%
Sequential Write BW (GB/sec) 5.3 2.1 152%
*Note: For almost all these maximum throughput benchmarks, all 32 Host Adapters must be utilized.
2.52x
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8700 Performance Matrix (Distributed workloads)
DS8K Metric DS8700 DS8300 % Change
"SPC-1-like" (IOs/sec) 155K 123K 26%
4K Read Hit (IOs/sec) 523K 425K 23%
4K Write Hit (IOs/sec) 203K 164K 23%
4K Read Miss (IOs/sec) 137K 111K 23%
4K Write Miss (IOs/sec) 109K 92K 18%
DBOpen 70/30/50 (IOs/sec) 191K 165K 14%
Sequential Read BW (GB/sec) 9,7 3,9 148%
Sequential Write BW (GB/sec) 4,7 2,2 113%
Single HA Read BW (MB/sec) 530 472 12%
Single HA Write BW (MB/sec) 400 265 50%
One DA Pair Read BW (MB/sec) 1239 866 43%
One DA Pair Write BW (MB/sec) 751 630 19%
2.48x
* Preliminary Pre-GA benchmarks
© 2011 IBM Corporation
IBM®
System Storage®
DS8700 Easy Tiering (5.1)
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
The trend towards larger disks continues!
DS8000 Drive Sizes Shipped by Quarter (percent)
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
SSDs compared to spinning disks
Disadvantages
� Wear issues– Errors may creep in over time– SSD controller designed in multiple
layers of error recovery (ECC, RAID, write-reduction, wear-leveling technology)
� Capacity– Relatively small capacity per drive unit
today compared to spinning disks– NAND and SSD industry expecting
breakthrough in capacity (SLC, MLC, XLC and PCM technology.)
� Cost– Significant more raw cost/GB today than
HDD– Need new technology to improve cost
performance ratio.
Advantages
� Absence of mechanical moving parts– Significantly more reliable
� Green– Power - Fraction of power consumed– Heat - Fraction of heat dissipated
and tolerant of wider range of temperature.
� Performance– Up to 100x more throughput and 10x
lower response time than 15K RPM spinning disks
– $/IOPS is significant better than traditional spinning disk.
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Multiple Tiers Storage view - Today
� Multiple Tiers Storage System manages different device types. Each device type may have hundreds to thousands of each device types.
� Optimize for – Cost Performance – Cost Capacity– Cost Reliability– Cost Power
Consumption– Scalability
43
IBM System Storage DS8000
SSD RAID Array(s)
FC/SAS RAID Array(s)
SATA RAID Array(s)
Applications
VirtualDisk
Virtual Disk Layer Managed Disk LayerStorage
Virtualization
Two Problems ! Little insights in application performance which can be affected by SSD characteristics Additional Storage Management for more tiers
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
44
Workload Learning through Smart Monitoring
� Each workload has its unique I/O access patterns and characteristics over time.
� Smart Monitoring and analysis tool allows customers to develop new insight to application optimization on storage infrastructure.
� This workload shows performance skews in three LBA ranges.
Inactive Disk RegionInactive Disk Region
Cool Data RegionCool Data Region
Lo
gic
al
Blo
ck A
dd
ress (
LB
A)
Lo
gic
al
Blo
ck A
dd
ress (
LB
A)
Time (Minutes)Time (Minutes)
Hot Data RegionHot Data Region
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
45
Demonstration of IBM Automated Data Reallocation
� Result:
– Improve application response time of 60-70+% at peak load by moving 5% of hot spot to SSD.
•Sustainability test, 76%
•Ramp test, 77%
� Demonstration of ADR:
– Compare “SPC-1 like”workload on HDD versus ADR with HDD and SSD”
– ADR Technology identifies and non-disruptively migrates “hot data” from HDD to SSD. About 4% of data is migrated from HDD to SSD.
60-70%+ Reduction in "SPC-1 Like" Average Response Time
with DS8000 Smart Tiering Technology
0
2
4
6
8
10
0 5000 10000 15000
IOPS
Re
sp
on
se
Tim
e (
ms
)
0.00%
20.00%
40.00%
60.00%
80.00%
100.00%HDD only
HDD+SSDwith SmartTiering
% Avg. RTReduction
ADRHDD+SSD
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
46
Traditional disk mapping Dynamic Virtual Storage mapping
All volumes appear to be “logically”homogenous to apps. But data is placed at the right tier of storage based on its usage through smart data placement and migration.
Volumes have different characteristics. Applications need to place them on correct tiers of storage based on usage.
Host
Volu
mes
DynamicVirtual volumes
Phys
ica
l D
evic
es
EASY TIER
Traditional Disk vs. Dynamic Virtual Storage Mapping
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Easy Tier - Overview
� Easy Tier is a new DS8700 feature that supports online dynamic relocation of data – Workload learning algorithms collect I/O statistics and generate a heat map of data
to be optimized on SSD– Data can be relocated to/from SSD and FC/SATA
• Automatic Mode
Provides automated extent level relocation granularity
Requires use of a merged extent pool• Manual Mode
Provides online dynamic volume relocation capability
– CLI/GUI setup and management– Storage Tier Advisor Tool (STAT) for I/O analysis and projected benefit
� DS8700 Hardware feature– Easy Tier is a new LIC feature available with Release 5.1 – Supported by all server platforms with no additional software
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Easy Tier - Data Relocation Options
Automatic extent level data relocation enabled in a Merged Extent Pool
Automatic ModeExtent Level Data Relocation
SSD Rank Pools FC or SATA Rank Pools
Merged Pool (SSD/FC or SSD/SATA)
Manual ModeVolume Based Data Relocation
Manual Mode (Volume Level) and Automatic Mode (Extent Level)
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Easy Tier – Automatic Mode Extent Relocation
Logical Volume Mixed Technology Extent Pool
SSD Arrays
HDD Arrays
Extent Virtualization
Hot Extents Migrate Up
Cold Extents Migrate Down
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Easy Tier – Automatic Mode Setup
� Options for configuring extent pools for Automatic Mode– Create new extent pool with SSD + HDD ranks– Merge existing SDD and existing HDD extent pool– Add SSD rank to existing HDD extent pool– Add HDD rank to existing SSD extent pool if no space efficient capacity (virtual
capacity or repository) configured in the extent pool
� Options for configure logical volumes for Automatic Mode– Create new logical volume in mixed technology extent pool– Migrate standard logical volumes between homogeneous and mixed
technology extent pool– Merge existing homogeneous extent pool with existing logical volumes with
another extent pool to create a mixed technology extent pool
� Existing recommendations continue– Each extent pool utilize same RAID format, size and speed of HDDs
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Easy Tier – Automatic Mode User Interfaces
� Design point is for customer to not normally need or use any controls to manage Automatic Mode
� Easy Tier Automatic Mode– Enable / disable
• Stops automatic mode extent migration while disabled
� Easy Tier Automatic Mode monitoring– Disable / managed volumes / all volumes
• Ignores workload activity in plan generation statistics while disabled• Plan generation suppressed if insufficient monitoring history since last plan (16 hours)• Managed volumes is normal setting and monitors workload on volumes in managed
pools if the Easy Tier LIC feature is enabled• All volumes causes all logical volumes to be monitored (with or without LIC feature)
Can be used with analysis tools prior to ordering and enabling the LIC featureCan be used to analyze logical volumes in homogeneous extent pools as
candidates for mixed technology pool
� Analysis Tools– Can offload reports on extent monitoring and obtain SSD capacity planning
recommendations– Can engage IBM for extended analysis and consulting
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Easy Tier – Manual Mode Overview
� Easy Tier Manual Mode allows a user to perform the following actions:– Volume Migration
• User can change a logical volume’s storage technology by dynamically relocating between extent pools
• User can change a logical volume’s extent allocation algorithm (EAM) (e.g. can re-rotate extents within the target extent pool)
– Extent Pool Merge• User can merge two existing extent pools without moving data
Consolidate extent pools with equivalent disks
Merge extent pools with to create a mixed technology extent pool for Automatic Mode
© 2011 IBM Corporation
IBM®
System Storage®
DS8700 (Metro) Global Mirror function enhancements
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
PPRC-XD OVER
LONG
DISTANCE
CHAN EXT
FlashCopy
0
10
1
0
00
10
00
0
10
REMOTE
APPLICATION
HOSTS
0
10
1
0
00
10
0
00
1
0
Change
Recording
bitmap
Out of
Sync
bitmap
PPRC 'A'
PRIMARY
CHAN EXT
PRIMARY
APPLICATION
HOSTS
PPRC ‘B’Secondary and FC Source
FC ‘C’Target
Local Site Remote Site
� Principe d’e la propagation d’état représenté par une bit map et non pas de mouvements
� PPRC Extended Distance (PPRC-XD) du site local vers le site remote
� Des copies automatiques cohérentes sont crées sur le site remote– Volume A utilise la “change recording bitmap” pendant la création du groupe de cohérence– Toutes les données sont envoyées sur le site remote et l”inband FlashCopy” sur le site remote crée un jeu de données cohérentes
sur le volume C
� Prévu pour s’ajuster sur les pointes sans impact sur les applications du primaires
Global Mirror : Principle
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Metro Global Mirror
C
D
B Global Mirro
r
A
Metro mirror
Site 1
Site 3
Site 2
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Local Site Remote Site
Master
Subordinate
GlobalMirror
Application1 Application2
Global Mirror – Single Session
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Local Site Remote Site
Master
Subordinate
GlobalMirror
Sessions can be divided by server or application.
Application1 Application2
Global Mirror – Multiple Sessions
© 2011 IBM Corporation
IBM®
System Storage®
DS8800Description
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8800 : Performance et efficiency
Tiroirs de disques
Batteries
Console
ContrôleursPOWER 6+
Rear
I/O drawers
Power Supplies
•New disk format•SFF disks 2,5”•6 Gb/s SAS (SAS-2)•24 disks per drawer
•New processors•IBM POWER6+
•New adapters •8 Gb/s FC•8 Gb/s Disks attachements
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
From DS8700 to DS8800
� Upgraded P6 Servers
– Leverage 2 or 4-way P6+ 5Ghz
� New host adapter
– 4 or 8 ports, 8Gb SAN connection, multi-core,
PCI-e
� New RAID adapter
– 8 ports, 8Gb FC, multi-core, PCI-e
– Better support for SSD modules
� New disk enclosure & power system
– Higher density, SFF SAS 2.0, 8Gb FC Fabric
– Hot aisle / cold aisle cooling design
P6
P6 P6
RIO RIO
New 2U disk enclosure for 24 SAS/SFF
HDD/SSD modules
DS8800
P6+
5GHz server
P6
Disks
New 8 Gbpshost adapter
New 8Gbps device adapter (up to 3,000
HDD capability)
• Reduced power requirements• Reduced floor space & service
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
61
DS8800: setting new environmental standards
DS8300/87001024 Drives in 5 Frames
DS88001056 Drives in 3 Frames
• ~40% less power
• ~33% less weight
• ~40% less floor space
DS8700 1-Frame128 Drives
DS8800 1-Frame240 Drives
• Same power consumption
• Same Weight
• Up to 87% more disks
240 336 480
128
128
240
256 256 256 128
480
GA
+ 6 months
480
GA
+ 12 months
Over 2000
Drives in
5 Frames(*)
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
DS8000 : performance leadership
units
DS8300
P5+
2.2 GHz 4way
DS8700
P6
4.7 GHz
DS8800(*)
P6+
5 GHz
Read Seq GB/s 4 9.7 12.5
Write Seq GB/s 2 5.7 6,4
DB open K IO/s 155 191 221
x 3.1
x 3.2
x 1.4
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
New Device Adapters + SSD : random accesses
0
10
20
30
40
50
KIO
/se
c
Read Write
Single DA
DS8800 DS8700 DS8300
0
20
40
60
80
KIO
/se
c
Read Write
DA Pair
DS8800 DS8700 DS8300
(*) DS8800 performance figures are preliminary, final numbers will be published at GA
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
0
500
1,000
1,500
2,000
MB
/se
c
Read Write
Single DA
DS8800 DS8700 DS8300
News Device Adapters : Sequential access
0
500
1,000
1,500
2,000
2,500
3,000
3,500
MB
/se
c
Read Write
DA Pair
DS8800 DS8700 DS8300
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Size Drive Type Drive Speed
73GB SSD (FC/LFF)
146GB SSD (FC/LFF)
600GB SSD (FC/LFF)
73GB FC (LFF) 15K RPM
146GB FC (LFF) 15K RPM
300GB FC (LFF) 15K RPM
450GB FC (LFF) 15K RPM
600GB FC (LFF) 15K RPM
1TB SATA (LFF) 7.2K RPM
2TB SATA (LFF) 7.2K RPM
Drive support in DS8700
Drive support in DS8800
10K RPMSAS (SFF)
FDE version also
600GB
Size Drive Type Drive Speed
300GB SSD (SAS/SFF)
146GB SAS (SFF) 15K RPM
450GB SAS (SFF)
FDE version also
10K RPM
1500GB NL-SAS (SFF) 7.2K RPM
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Hard Disk Drives: 3.5” vs. 2.5”
0
500
1000
1500
2000
IOPS
Random read Random write
DS8000 6+P array performance
3.5" 450GB 15K 2.5" 450GB 10K
0
200
400
600
800M
Bps
Seq read Seq write
DS8000 6+P array performance
3.5" 450GB 15K 2.5" 450GB 10K
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Code Streams in 2010 and 2011
Release 5.0
Release 5.1
Release 6.0
DS8700w/LFF drives
DS8700 w/LFF drives• Easy Tier• …
DS8800 w/SFF drives
Release 6.1
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
R6.1 Functions (1H2011)
� Thin Provisioning (distributed only) , � Quick Initialization (distributed only)� High Performance FICON (Multitrack)� Easy Tier� Eight drive install groups of Solid State Disk (SSD) � Remote Pair FlashCopy®� Prevent deletion of CKD and FB volumes in use � Disable recovery key � Rekey data keys of an encryption group � z/HPF extended distance� Multiple Global Mirror sessions (RPQ only)
*** IBM Confidential ***
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Current DS8700 roadmap
� DS8700 models will coexist with DS8800
� Additional 600GB SSD choice in 4Q2010
� Enhanced 8Gbps Host Adapter in 1H2011
� Additional code enhancements in R6.1
� More enhancements TBD
*** IBM Confidential ***
All statements regarding IBM future direction and intent are subject to change or withdrawal without notice, and represents goals and objectives only.
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011
Thank You!
© 2011 IBM Corporation
Journée Technologique STG IBM – Bois Colombes, 13 Janvier 2011