Post on 15-Feb-2018
FlashSystem family 2015
IBM FlashSystem®
Portfolio
Flash Realized. It’s about time.
Peter Frankcam
pete@uk.ibm.com
07710 241012
@PJFTech
Technical Specialist
IBM FlashSystems
© 2015 IBM Corporation 2
IBM’s flash journey so far
• IBM Acquires
Texas Memory
Systems (TMS)
• 19th TMS
RamSan
product
generation
• $1B IBM Flash
Investment
• FlashSystem
820/720,
810/710,
• #1 Market
Share*
• FlashSystem
V9000
• FlashSystem
900
• More coming
soon!
• FlashSystem
V840
• FlashSystem
840
• IBM Data
Engine for
NoSQL (CAPI)
Third Anniversary of Texas Memory Systems Acquisition
* Source: Gartner #258537, Market Share Analysis: SSDs and Solid-State Arrays, Worldwide, 2013, June 2014
© 2015 IBM Corporation 3
*Calculations by IBM based on Gartner Report: Market Share
Analysis: SSDs and Solid-State Arrays, Worldwide, 2013
**Source: IDC, 2014
Shipped more capacity in 2014 than the two closest competitors combined**
Global market acceptance #1 with 25% market share
Clear leadership position 44% greater revenue than next closest competitor*
IBM Ranked #1 for Flash Storage(SSA) Market Share, Worldwide based on
revenue for 2013 and shipment data for 2014
What have we achieved?
© 2015 IBM Corporation 4
How did we
get here?
© 2015 IBM Corporation 4
© 2015 IBM Corporation 5
We’ve arrived here by way of 3 technology
transformations and one epic problem
• Processor performance increases have followed
Moore’s law
• Unprecedented growth in concurrent users and
transactions
• Rapid adoption of flash in consumer electronics
drove up flash density and drove down flash price
• Disk drives increased capacity at the expense of
performance/capacity
This demands an revolutionary solution,
the next generation IBM FlashSystem
© 2015 IBM Corporation 6
Performance
Gap
CPU performance up 10x this last decade
Storage has grown capacity but unable to keep up
in performance
Systems are now Latency & IO bound resulting
in significant performance gap
IT infrastructure challenges
© 2015 IBM Corporation 7
Most Costly
& Volatile
Time
Consuming,
Very
Expensive
& Risky
Wasteful,
Expensive &
Ineffective
with Storage
Latency
Issues
Expensive
& Ineffective
for Storage
Performance
Issues
Client responses to performance gap
Add More
Memory
Typical Performance
Mitigation Tactics
HDD Performance Enhancement
Add CPUs
Tune & Modify
Application
© 2015 IBM Corporation 8
Introducing IBM FlashCore
© 2015 IBM Corporation 8
™
© 2015 IBM Corporation 9
Introducing IBM FlashCore™
IBM uses tools and technologies to drive a robust solution…
Industry Standard Interface Technology
Powered by
IBM FlashCore™ Technology
IBM MicroLatency™ Module
Hardware Accelerated I/O
Advanced Flash Management
Industry Standard Chip Technology IBM Enhanced
Micron Technology
© 2015 IBM Corporation 10
IBM MicroLatency™ Module SSD
IBM MicroLatency™ Module vs SSD
CUSTOM
Requires hardware engineering
FAST
Designed for low latency
RELIABLE
Innovative flash management
COMMODITY
No hardware skills needed
SLOW
Designed for low cost
FALLIBLE
More single points of failure
Better Performance Better Reliability Better Density Lower $/IOPS
© 2015 IBM Corporation 11
FlashCore: Hardware Accelerated I/O
• Engineered for Flash
• Hardware RAID
• Non-blocking Crossbar Switch
• Hardware Only Data Path
• Single Box Highly Available
Architecture
• Concurrent Code Load
• Concurrent Maintenance
Canister-1 Canister-2
FC-1
MC-2MC-1
FC-3 FC-4FC-2
XBAR 0 XBAR 1
© 2015 IBM Corporation 12
FlashCore: IBM MicroLatency™ Module
• IBM Engineered
• Massively Parallel Design
• FPGAs in the Data Path
• Distributed RAM
• High Speed Interface
• Hardware-based Data at Rest Encryption
© 2015 IBM Corporation 13
IBM Advanced Flash Management improves MLC flash endurance 9x
over standard implementations without sacrificing latency.
FlashCore: Advanced Flash Management
“Through close collaboration between IBM research, development and flash engineering, IBM created an
advanced flash characterization platform to test thousands of flash devices over a lifetime of wear. As a
result, IBM has made major strides in advanced flash management.” – Andy Walls, IBM Fellow
Key accomplishments:
• Innovative ECC algorithm corrects very high bit error rates
• Variable voltage and read level shifting maximizes flash endurance
• Advanced Flash Management continually monitors the health of blocks
and performs asymmetrical wear leveling and sub-chip tiering:
- Health binning: up to 57% improvement in endurance!
- Heat segregation: up to 45% reduction in write amplification!
• Micron Technology provides special access to the chip
Backed by IBM’s Flash Wear Guarantee!
9.4X Endurance Improvement
Standard MLC Flash
IBM Enhanced MLC
© 2015 IBM Corporation 14
Powered by
IBM FlashCore™ Technology
Extreme Performance with IBM
MicroLatency™: Delivers 100 microsecond
response times
Macro Efficiency: Lowest latency offering
with >40% greater capacity
Enterprise Reliability: IBM enhanced
Micron flash technology which contains MLC
offering with Flash Wear Guarantee
Scalable Performance: Grow capacity and
performance with up to 2.2 PB scaling capability
Enduring Economics: Next generation
flash media for less than the cost of HDD
Agile Integration: Fully integrated system
management to simplify management and
improve workforce productivity under a single
name space
IBM FlashSystem family
IBM FlashSystem 900 IBM FlashSystem V9000
© 2015 IBM Corporation 15
FlashSystem 900
© 2015 IBM Corporation 15
© 2015 IBM Corporation 16
• IBM MicroLatency™ with up to 1.1 million IOPS
• 40% greater capacity compared to previous generation
• IBM FlashCore™ technology, our secret sauce
Technical collaboration with Micron Technology – flash chips
• IBM enhanced flash technology
• MLC NAND flash offering and Flash Wear Guarantee
Improved integration with VMware VASA and VAAI UNMAP,
OpenStack Cinder
Performance at-a-glance
Introducing IBM’s premier all-flash offering IBM FlashSystem 900 the next generation in our lowest latency offering
Minimum latency
Write 90µs
Read 155µs
Maximum IOPS 4 KB
Read (100%, random) 1,100,000
Read/write (70%/30%, random) 800,000
Write (100%, random) 600,000
Maximum bandwidth 256 KB
Read (100%, sequential) 10GB/s
Write (100%, sequential) 4.5GB/s
Module type 1.2TB 2.9TB 5.7TB
Modules 4 6 8 10 12 6 8 10 12 6 8 10 12
RAID 5 capacity 2.4 4.8 7.2 9.6 12 11.6 17.4 23.2 29.0 22.8 34.2 45.6 57.0
© 2015 IBM Corporation 17
Extreme Performance enable business to unleash the power of performance, scale, and insight
to drive services and products to market faster
• Improved end-user experience
• Faster insights into critical applications
The value of performance
A
1-SECOND
DELAY
IN PAGE
LOAD TIME
=
7% LOSS
IN
CONVERSIONS
11% FEWER
PAGE VIEWS
16% DECREASE
IN CUSTOMER
SATISFACTION
In dollar terms, this means that if your site typically earns
$100,000 a day, this year you could lose $2.5 million in sales. Source: Aberdeen Group
© 2015 IBM Corporation 18
97 percent reduction in processing time
without replacing a single server
20x more data crunched within the existing
window and SLAs
150 million cases of product per week
Solution Components • SAP ERP
• JDA Manugistics
• IBM FlashSystem 820
• IBM SAN Volume Controller
Coca-Cola Bottling Co. Consolidated Delivering deeper insight into customer demand four times faster
Business challenge: CCBCC needed to process more data to accurately
anticipate customer demand without increasing time-to-insight or exceeding SLAs.
The solution: CCBCC now processes 20 times more forecasting data, down to
the store level, which enables them to match manufacturing output with demand,
enables earlier logistics planning and increases profitability.
“By using IBM FlashSystem to accelerate our insights into customer
demand, we’re better placed than ever before to offer unbeatable levels
of service to our customers across the United States.”
- Tom DeJuneas, Infrastructure Manager, CCBCC
Video Case Study
© 2015 IBM Corporation 19
Macro Efficiency
driven by consolidation of hardware and
software, deployment speed, efficient use of IT
staff as well as power and cooling savings
Source: IBM Client Experiences
The data below are based on average operating conditions that may or may not be representative of a
particular customer’s operating environment. The use case measurements are from IBM Flash customers
using the flash technology that has been integrated into IBM’s systems.
No changes to application or architecture
Benefits & economics outweigh disk
Reduce floor space, power & cooling
Boost IT efficiency
98% Reduction
in application processing time
50% Reduction
in TCO
97% Reduction
in physical footprint
95% Reduction
in power consumption
© 2015 IBM Corporation 20
Maximize business uptime
• Superior protection with multiple RAID layers
• Advanced wear leveling and over-provisioned space
• Non disruptive maintenance and current code load
• Hot swap & redundant components
Superior Protection: Beyond Disk RAID
Variable Stripe RAID
Two-dimensional flash RAID
Variable voltage
Read level shifting
High-speed clock recovery
Heat segregation
Health binning
Advanced Engineering =
Less Maintenance
Highest reliability levels
Purpose-built, Enterprise Architecture
Enterprise Reliability
durable and reliable designs that use
enterprise class flash and patented data
protection technology
VSR within modules
2D RAID
across
modules
© 2015 IBM Corporation 21
FlashSystem V9000
© 2015 IBM Corporation 21
© 2015 IBM Corporation 22
IBM’s premier, fully-integrated, Tier 1, all-flash offering
© 2015 IBM Corporation 23
IBM FlashSystem V9000
IBM introduces a fully integrated, flexible, feature rich all-flash storage system
Introducing IBM’s choice for open system tier 1 storage
General availability: April 24
• Scalable all-flash architecture accelerates applications and entire
infrastructures
• Performs at up to 2.5M IOPS with IBM MicroLatency
• Up to 57TB usable (285TB effective capacity) in only 6U and scales to
456TB usable (2.28PB effective capacity) in only 34U
• New licensing structure to simplify ordering and planning for External Data
Virtualization, FlashCopy, Metro Mirror, and Real-time Compression
• FlashSystem Tier 1 Guarantee
Scalable Performance Agile Integration Enduring Economics
Powered by IBM FlashCore™
Technology
© 2015 IBM Corporation 24
• Scalable to 2.2 PB effective capacity
• Flash for less than the cost of disk
with IBM Real-time Compression
• Low power and cooling requirements
• Virtualize storage
• Flash wear warranty
• Infrastructure continuity with space
efficient snapshots, cloning and
replication
• Fully integrated system management
• Application-aware data services
• AES, data at rest encryption
• Tier or mirror to existing storage
• Mixed workload consolidation
• Non-disruptive data migrations
• Concurrent code load
• Scale-up or scale-out, independently
• Scalable to 2.5 million IOPS
• Scalable to 19.2 GB/s bandwidth
• Sustained IBM MicroLatency™
• Quality of Service
FlashSystem V9000: paradigm shift in enterprise storage Dimensions of Value
Scalable Performance
Agile Integration
Enduring Economics
© 2015 IBM Corporation 25
FlashSystem V9000 makes applications faster Scalable Performance
Scalable Performance
CPU Utilization
& App. Efficiency
4%
Total Application
Processing Time
5,200μs (5.2ms)
5,000μs (5ms)
200μs (.2ms)
Application
Processing Time
Time Waiting for I/0 (Waiting for Array)
Time Processing Data (Server CPU)
Disk/Hybrid
SW-Based SSD
HW-Based
Solutions
CPU Utilization
& App. Efficiency
50%
Total Application
Processing Time
400μs (.4ms)
200μs (.2ms)
200μs (.2ms)
What will you do with
the Extra Time?
© 2015 IBM Corporation 26
Maximum Performance
Scaled Out (100% Read, 4 Building
Blocks) *
Latency (4K) 200µs
IOPS (4K) 2,520,000
Bandwidth (128K) 19.2 GB/s
Maximum Performance
Scalable Build Block (100% Read, Cache Miss)
Latency (4K) 200µs
IOPS (4K) 630,000
Bandwidth (128K) 6.2 GB/s
Maximum Performance
Fixed Build Block (100% Read, Cache Miss)
Latency (4K) 200µs
IOPS (4K) 526,000
Bandwidth (128K) 6.2 GB/s
Two-dimensional scaling Scalable Performance
Scalable Performance
© 2015 IBM Corporation 27
FlashSystem V9000 options for deployment Scalable Performance
• Great for multiple mixed
workloads that drive huge I/O
• Scale out for more all flash
capacity, IOPS and bandwidth
• Up to 2.5M IOPS, 200µs
• Up to 228 TB usable, 1.1 PB
Effective
Mixed Workload Accelerator
• Great for large data sets with
big I/O requirements and
needing storage services
• Scale up for more all flash
capacity
• Up to 630K IOPS, 200µs
• Up to 285 TB usable, 1.4 PB
Effective
Application Accelerator
• Great for Tier 1 disk
replacement
• Scale up and out for
more processing, more
capacity and more I/O
• Up to 2.5M IOPS,
200µs
• Up to 456 TB usable,
2.2 PB Effective
Public or Private Cloud
• Great for data centers with
heterogeneous storage
• Extends core feature set to
other storage arrays
• Up to 2.5M IOPS, 200µs
• Up to 3 Exabytes virtualized
Virtualized Data Center
• Great for database and VDI
workloads
• Accelerate, migrate, tier,
clone, snapshot, replicate,
compress existing storage
• Up to 630K IOPS, 200µs
• Up to 57 TB usable, 285 TB
effective
Small Data Center
Scalable Performance
© 2015 IBM Corporation 28
• Improve workforce productivity
• Simplify management
• Single name space
Fully integrated system management Agile Integration Agile
Integration
© 2015 IBM Corporation 29
Flexible volumes Agile Integration
• Generic – fully allocated
• Thin-Provision – space consumed on-demand
• Mirror – two copies of volume
– Migrate between pools or change
volume type
– Preferred read mirror between flash
and legacy storage (all reads from flash)
• Thin mirror – a mirror with copies being thin
• Compressed – volumes utilizing Real-time Compression
Agile Integration
© 2015 IBM Corporation 30
Non-disruptive data migrations Agile Integration
Non-disruptive data migration
• Maintain concurrent I/O access
• Externally virtualize the existing
storage system
• Move data off older storage
onto flash storage
Agile Integration
© 2015 IBM Corporation 31
Flash for less than the cost of Tier 1 disk Enduring Economics
Database Office Files CAD/CAM Geoseismic Virtual Servers
JPEG/ZIP
80% 30% 5% 50% 72% 70% Compression Rate
Traditional
Storage
Average of
65% Compression
Real-time Compression Efficient
Half the Data Storage
Real-time Compression, for active and inactive data
Enduring Economics
© 2015 IBM Corporation 32
Storage virtualization Enduring Economics
Preserve your existing
investment in storage
Manage Tier Protect Compress with Real-time Compression
Optimized performance at lower overall cost IBM Easy Tier® flash storage management
Thin Provisioning
More productive use of available storage
Across all supported host platformsWithout thin provisioning, pre-allocated
space is reserved whether the
application uses it or not.
With thin provisioning, applications can
grow dynamically, but only consume space
they are actually using.
Dynamic
growth
Busiest data extents are identified and automatically relocated to highest performing Solid-state Disks
Remaining data extents can take advantage of higher capacity, price optimized disksHot-spots due to poor data layout. Optimized performance and throughput.
AutomaticRelocation
SSDs HDDs SSDs HDDs
IBM Easy Tier SSD Management
Mirror data off-site
Synchronously over Metro distances.
Asynchronously over Global distances.
Application-level consistency groups.Network
Flash Flash HDDs HDDs
Enduring Economics
© 2015 IBM Corporation 33
Metro Mirror – Synchronous
– Updates are committed to both sites
before acknowledged to server
– Supported up to 300 km, adds
additional latency to hosts
Global Mirror – Asynchronous
– Server update acknowledged in
parallel to update to remote site
– Supported up to 80ms of round-trip
latency, distance dependent upon
network quality
FlashCopy Options
• Full or “thick”
• Thin nocopy
• Incremental
• Cascading
• Multi-target
• Consistency groups
• Reverse
Infrastructure continuity Enduring Economics
• Snapshot – Instant space-efficient,
thin provisioned copy, capture state of
volume before a change
• Clone – Creates a complete one-time
replica of source, testing data or
attaching to other servers
• Backup – Complete point in time
replica that can be incrementally
refreshed
Full set of disaster recovery tools:
Enduring Economics
© 2015 IBM Corporation 34
IBM FlashSystem tier 1 guarantee
Data Reduction: The eligible machine, using compression, will achieve a system level
storage efficiency savings rate of at least 3:1, based on Comprestimator performance
modeling results, within the guarantee period.
Performance: IBM MicroLatency™ performance.
Endurance: Flash memory will be covered for read/write endurance as
long as you are under warranty or maintenance.
7 year 24 x 7 Support: Up to 7 years support available with optional price
protection and flash media retention offerings.
Peace of Mind: No Charge, Complimentary IBM services for all Tier 1
opportunities: 40 hours for 228+ TB solutions up to 556 TB, or 80 hours for
556+ TB solutions, Lab Services delivery in 2015.
IBM FlashSystem – GUARANTEED!!!
© 2015 IBM Corporation 35
Improve Performance
Oracle • Standalone
• RAC
MySQL
SQL Server • Standalone
• Clustered
• Always-On
DB2
Sybase
SAP HANA
Flash is an equal opportunity database accelerator.
You name it, flash makes it faster.
Industry Leading Databases Accelerated with Flash
© 2015 IBM Corporation 36
Introducing game-changing innovation Accelerates big data & analytics
IBM Data Engine for NoSQL – Power Systems
Edition
Integrated platform for large and fast growing NoSQL data stores
which utilizes high speed access to both RAM and flash storage for
lower cost and higher workload density than a standard RAM based
system – with equal or better performance than scale out.
Can be offered as a cloud-based service or as an on premise
solution for enterprises.
The Value: Lower Total Cost of Ownership for NoSQL infrastructure with high levels of
consolidation and no performance impact or degradation.
3X Cost Savings
24:1 physical
server consolidation
6x less rack space 2U server+2U FlashSystem vs. typical deployment
CAPI
• Direct attach IBM FlashSystem® to POWER8 via Coherence
Accelerator Processor Interface (CAPI)
• Issues Read/Write commands from applications, like redis, to
eliminate 97% of code path length
• Saves 20-30 cores per 1M IOPs
Direct
attach
via FC
For more information, please visit IBM FlashSystem EcoSystem
© 2015 IBM Corporation 37
IBM FlashSystem® improves DB2 workloads
Transactional Workload Brokerage Workload Big data insights analytics workload
40X better TPS
performance
190X better TPM
performance
86X more IOPS
95% price-per-performance
improvement
90% price-per-performance
improvement
7X performance improvement
for intermediate queries and
2.5X performance improvement
for complex queries.
40X better IOPS
IBM® Power Systems™ provide the processing capabilities
needed to scale out or scale up while powering the most
complex analytics.
These IT infrastructure components provide the foundation
upon which to build a next generation data platform—using
IBM DB2®—that keeps transactional workflows and analytics
operating at maximum efficiency.
IBM FlashSystem® storage offers performance, efficiency and
reliability, as well as the virtualization tools that enable
enterprises to turn the explosion of data into insight for
business advantage.
How can the growing data volumes between
different processing clusters, departments and
sites be shared most effectively? How can the
explosion of unstructured data be managed?
For more information, please visit IBM FlashSystem EcoSystem
© Copyright IBM Corporation, 2015
An IBM FlashSystems Assessment tool for IBM i workloads FLiP - Flash IBM i Performance
© 2015 IBM Corporation 39
The best IBM i jobs candidates are the ones spending
proportionally most of their total run time waiting for Read
disk I/Os.
This tool will help you identify
those jobs that potentially
could take benefit from IBM
FlashSystems technology in
order to improve their runtime
performance.
© 2015 IBM Corporation 40
Disk READ
time
Disk WRITE
time
Other Disk
contention
time
CPU time Average Disk
READ resp.
time (msec)
Average Disk
WRITE resp.
time (msec)
Disk Read Time
vs CPU time
Idle Time Other
Wait Time
Total
Time
Disk Read vs
Total (%)
Number of
READ I/Os
Number of
WRITE I/Os
52024.52 1.15 2.51 2581.94 5.469 0.703 74 0 202.37 54812.52 94.9% 9512620 1636
13037.47 2.62 1.1 2138.11 5.949 0.825 18 0 3178.48 18357.79 71.0% 2191540 3176
12357.67 1.08 0.65 531.98 4.348 0.73 66 0 0.01 12891.41 95.9% 2842150 1479
8524.67 0.06 0.01 339.9 3.514 1.512 89 0 0.01 8864.66 96.2% 2425916 40
8008.73 1.05 2.9 1175.2 5.916 0.644 15 0 0.02 9187.92 87.2% 1353741 1630
7011.81 0.3 0.02 1342.87 6.733 0.796 16 0 67.61 8422.63 83.2% 1041409 377
5317.69 0.73 0.16 960.21 8.118 0.985 18 0 88.4 6367.2 83.5% 655049 741
5010.22 0.54 0.09 1930.82 6.219 0.835 8 0 211.38 7153.07 70.0% 805631 647
4359.98 0.15 0 796.77 4.286 0.76 17 0 0.08 5157.01 84.5% 1017261 197
4163.4 1.22 0.54 620.55 7.077 0.627 20 0 971.15 5756.88 72.3% 588300 1946
3999.97 0.05 0.01 472.31 1.237 0.813 28 0 0.22 4472.58 89.4% 3233605 62
3853.8 11.08 0.58 1219.55 5.195 0.962 10 0 79.18 5164.22 74.6% 741829 11518
3462.64 17.74 0.21 752.27 6.481 1.134 14 0 63.23 4296.12 80.6% 534276 15644
3425.39 0.31 0.07 922.85 7.146 0.932 11 0 45.37 4394.01 78.0% 479344 333
3298.61 2.19 0.8 170.38 5.846 0.717 64 0 0.1 3472.1 95.0% 564251 3054
4.926 0.948 31 88.9%
4.926
3.327 (based on the weighted average resulting from EasyTier Skew Level) (below information can be retrieved from the IBM i performance reports for Disk Magic)
0.485 (see TAB "Flash Latency" for more information at the end of this document) Max number of READ IOPS : 20,000
36% (Depending on the Skew Level and the Fraction of Flash capacity selected below) READ Workload characteristics : 100% Reads 32 KB SVC
14% (see TAB "EasyTier Skew Level" at the end of this document) Corresponding Flash Latency (msec) :
19%
32% Workload Skew level (EasyTier) Very Low Very Low Max number of WRITE IOPS : 10,000
48% Fraction of capacity on Flash 20% 20% WRITE Workload characteristics : 100% Writes 32 KB SVC
1.3 Fraction of READ I/Os redirected to Flash 36% Corresponding Flash Latency (msec) :
Those are the only cells where you need to act before pushing the button, all other cells are protected and computed automatically. Selected Skew Level : Very Low
0.621
20,000
100% Reads 32 KB SVC
0.485
100% Writes 32 KB SVC
10,000
Generate a PowerPoint document with all the charts
The procedure - performance data created from the MGTCOL
40
1. Run the query provided 2. Filter the query result
3. Select what you want to model 4. Analyze the resulting charts
© 2015 IBM Corporation 41
FLiP - Wait times comparison
© 2015 IBM Corporation 42
FLiP - Current vs Projected Total Job Run Time (hour)
© 2015 IBM Corporation 43
CPU Efficiency & Read Latency Improvements
FlashSystem family 2015
Questions? Flash Realized. It’s about time.