Big Data Flash

Sales Introduction
IBM DeepFlash 150 and Spectrum Scale

Firat Ozturk
IBM MEA Flashsystem Technical Sales Leader
firat@ae.ibm.com
06.12.2016, Riyadh
 There is a significant shift in industry, driven by new workloads and
by Flash transformation
The market for flash-based arrays, which includes both
Over the next five years, more than 95% all-flash arrays (AFAs) and hybrid flash arrays (HFAs)
of all expenditure on information was worth over $11.3 billion in size in 2014 and will
technology (IT) will go toward 3rd exhibit strong double digit growth rates over the
Platform computing infrastructure. next five years.
IDC March 2015 IDC March 2015
80% of the Storage Purchases by 2017 Storage purchased are bifurcated towards Flash
will be associated with New Workloads* and large capacity disks

80%

2013 2014 2015 2016 2017

New Workload 11.5% CAGR The rest -13.5% CAGR
Source: IDC's Worldwide and U.S. Enterprise Storage Systems 2014-2018
Storage Opportunity: associated with Cloud, Analytics, Mobile and Forecast IDC's Storage User Demand Study, Fall 2014 Release
Social vs. the Rest
*Source: IBM Systems CAMS GMV

2 IBM and Business Partner Internal Use Only IBM Systems

 Why such a dramatic shift in spend?
Pursuit of the ever elusive mirage of sustainable competitive advantage

Excellence in operational skill used to confer long-term advantage,

Leaner manufacturing
Made higher-quality products
Had superior distribution
With this capabilities, you could outrun your competition, but today they are table
stakes!

The new source of competitive advantage is customer centricity: deeply

understanding your customers needs and fulfilling them better than anyone else.
Harvard Business Review, Building and Insights Engine, September 2016

3 IBM and Business Partner Internal Use Only IBM Systems

What matters now is not so much the quantity of
data a firm can amass but its the ability to
connect the dots and extract value from the
information. This capability differentiates successful
organizations from less successful organizations.

According to Insights2020 research, 67% of the

executives in overperforming firms said that their
firms were well skilled at linking disparate data
sources, whereas only 34% at underperforming
firms made the same claim.
Harvard Business Review, September 2016. Insights202 study conducted by strategic consultancy, Kantar
Vermeer, in cooperation with The Wharton School.
IBM Systems | 4
 New Class of Flash: Big Data Flash
Scalable capacity and performance at low price points for big data

Performance can Often do not

Petabyte Scale Written once but
of unstructured leads business benefit from
read often,
Big Data
data reduction
data and results, example
technology,
example: video Attribut
growing rapidly faster time to and image
insights example already es
compressed files

Performance consistently better than that of

the best HDDs today
Cost comparable to that of performance
optimized HDDs
Flash media that leverages flash Economics
Systems implementations that support
massive scalability and meet enterprise
Requirements
Source: IDC, 2015
Targeted primarily at big dataSource:
and secondary
IDC, 2015
storage environments

5 IBM and Business Partner Internal Use Only IBM Systems

 A well known and not so well known trends in the data storage world

Data growth continues But its not just data volume that is growing,
to be explosive driven all of the following are also growing:
by digital transformation of
almost every industry and
every business function

Data Volume Data Velocity Data Variety Data Value

Data is not coming in Data is the resource

Data continues How fast data
structured tables, its all on which a businesss
to double every is coming in and
unstructured- images, competitive advantage
9 months going out
log files, sensor data, depends
etc.

Its the growth in all of the above that is creating the demand for
Big Data applications and infrastructure

6 IBM and Business Partner Internal Use Only IBM Systems | 6

 What does Big Data storage look like today?
Scale out File Storage with lots of back-end HDDs
to provide high throughput and high capacity:

Scale Out File Server Scale Out File Server

Challenges with this infrastructure:
Inconsistent performance
Low throughput density
High latency
High power consumption
High floor space
High cooling costs
High disk failure rates
..

7 IBM and Business Partner Internal Use Only IBM Systems | 7

 New Big Data alternative: instead of HDD, use Big Data Flash
For clients who value application response time and/or throughput per rack unit

Move from Big Data

HDD configuration

To this Big Data Flash

Hard Drives
configuration

Increase application response time by 8X

All Flash
Increase throughput/rack unit by 2.8X
File Server 10U
25GB/S
28U File Server Reduce MTBF
25GB/S

Reduce power & cooling costs by 30%-50% All Flash

Hard Drives

8X faster response time

and same throughput
as the HDD version

8 IBM and Business Partner Internal Use Only IBM Systems | 8

 Not conventional Flash, a new class of Flash: Big Data Flash
Scalable capacity and performance at low price points for big data

HDD DeepFlash Conventional Flash

Price $ $$ $$$
Performance 10s of milliseconds Sub Milliseconds Micro Seconds
Attributes High ingest rate Extremely latency sensitive
Low change rate Can justify price premium
High read rate
Typical use Big Data analytics (ex: video, health VDI, Server Virtualization, Database
cases care data), Hadoop, Spark and Application Acceleration

Performance consistently better than that of the best HDDs

Cost comparable to that of performance optimized HDDs

Systems implementations that support massive scalability and meet enterprise

requirements

9 IBM and Business Partner Internal Use Only IBM Systems | 9

 IBM Extends Flash Leadership by Introducing IBM DeepFlashTM 150
Combined with IBM Spectrum Scale Software Defined Storage
Game changing
Higher Reliability performance
Up to 2M raw IOPS
sub 1ms latency
1.5+ million hours MTBF (mean time 12GB/s throughput
between failure)
Hot-swappable architecture - easy FRU of
fans, SAS expander boards, power
supplies, flash cards Solution with
Directly samples air temp Spectrum Scale
Higher Density 40 80 hours IBM Lab services
Scale-out storage software that
3U chassis starting at 128TB up to scales Exabyte's of DeepFlash
512TB arrays under a single name
8 to 64 8TB flash cards space
Unified data access including,
file, object, HDFS and OpenStack
150W(idle), 750W(abs -max), typical Seamlessly tiering across Flash,
workload 450W Disk, Tape and Cloud
30% to 50% lower power consumption and Efficient Space-Saving
cooling power requirement of a equivalent Compression
HDD arrayLower Energy Low overhead encryption
Consumption

10 IBM and Business Partner Internal Use Only IBM Systems

 DeepFlash 150 overview

3U x 30 rack mount storage enclosure

64 internal cards, 8TB each
Presented as 64 SCSI targets - JBOF behind
redundant SAS expanders
8x SAS 12Gb/s connectors (4-lane)
Redundant expanders, power supplies and fans, with
hot-swap Flash Modules
Hot-swappable architecture - easy FRU of fans, SAS
expander boards, power supplies, flash cards
150W(idle), 750W(abs -max), typical workload 450W
30% to 50% lower power and cooling power
requirement of a equivalent HDD array

11 IBM and Business Partner Internal Use Only IBM Systems | 11

 Inside the IBM DeepFlash 150

Storage elements are called a Board Solid

State Drive (BSSD)
The BSSD is a cost-optimized implementation of
an SSD in a different form factor.
Each BSSD provides 8TB of raw flash, with
the maximum 64 BSSDs in 3U chassis
delivering a total capacity of 512TB.
Standalone DeepFlash 150 enclosures are
BSSD available in three sizes: 128TB, 256TB and
512TB.
External connectivity with DeepFlash 150 is
12 Gb/s SAS (SAS 3.0).
Each DeepFlash 150 enclosure has two Host
SAS Expanders (HSEs)
Each with four mini-SAS connections for
HSE interfacing to:
One or more external servers through SAS 6Gbit
links using SAS cables

12 IBM and Business Partner Internal Use Only IBM Systems | 12

 IBM
DeepFlash 150
storage
enclosure

13 IBM and Business Partner Internal Use Only IBM Systems | 13

IBM FlashSystem 900

IBM
DeepFlash
8TB BSSD

(Board Solid State

Drive)
14 IBM and Business Partner Internal Use Only IBM Systems | 14
 Deep Flash 150 hardware

IBM DeepFlash 150 provides an

Small Medium Large
essential big-data building block for
petabyte-scale storage MTM 9847-IF1 9847-IF2 9847-IF3
environments. Flash cards 8TB

When bundled with Spectrum Raw Capacity 128TB 256TB 512TB

ScaleTM it is an ideal choice to Usable 123TB 245TB 490TB

Capacity
accelerate systems of
Volume count 16 32 64
engagement, unstructured data,
big data and other workloads MES 128TB Upgrade Kit 128TB Upgrade,
16 x 8TB
requiring low latency
HW Warranty 1 or 3 years
Power Typical full load, 512TB = 450 watts
consumption Idle, no load = 150-200 watts

15 IBM and Business Partner Internal Use Only IBM Systems

Spectrum Scale + Standalone DF150 Building Block
reference architecture
You can build your own Spectrum Scale +
DeepFlash 150 building block - should consist
of:
Two Spectrum Scale IO servers
Connected to two DeepFlash 150 enclosures
Protection from failure by:
Spectrum Scale replicating data / metadata
Across 2 DF150 JBOFs (Just a Bunch of Flash)

DF150s connected by SAS cables to

Spectrum Scale IO servers
Quorum node (can be a virtual machine) is
required to maintain Spectrum Scale quorum
Automatic failover uses a third failure group
(only 100MiB in size) as
Defined on a disk partition local to one other
cluster nodes

16 IBM and Business Partner Internal Use Only IBM Systems

 Our selection took into account:
Product Experiences End to End protection for data flow without data
from Silicon Valley loss

Integration with a software stack that would

enhance scalability
Spectrum Scale for Global NameSpace
Spectrum Scale Spectrum Scale

Advanced functions
Hadoop transparency layer
AFM (Spectrum Scales Active File
Management)
Object Storage
IBM DeepFlash 150 building block

Policy driven events

Tier to and from storage with different
characteristics and data classification
17 IBM and Business Partner Internal Use Only IBM Systems
 The Offering: IBM DeepFlashTM 150 with Spectrum Scale
IBM DeepFlash 150 Small Medium Large
The flash, path or enclosure failure is protected by 9847-
Spectrum Scale by replicating both the data and HW MTM 9847-IF2 9847-IF3
IF1
metadata across two failure groups
Raw Capacity 128TB 256TB 512TB
Typical configuration HW List Price (1 year $552,50 $1,075,00 $2,093,87
warranty) 0 0 5
Example HW Net price
Spectrum Scale Spectrum Scale (75% max green discount) $138,12
NSD Server 1 NSD Server 2 $268,750 $523,469
Note: no yellow zone; >75% 5
Quorum-Manager Quorum-Manager
red zone
Example max green HW
$1.079 $1.050 $1.022
$/GB raw Current clients may use existing socket based
Spectrum
Example HW $/GB if Scale License (e.g. Machtype 5725-
replicated Q01 partno. D148KLL $2.16 ) $2.10 $2.04
Spectrum New clients,
(2x IBM DeepFlash 150) encourage to sell the new Spectrum
Scale Scale capacity based license (socket based
Spectrum Scale Spectrum Scale Spectrum Scale Capacity
license option still available)
Failure Group 1 Failure Group 2 based License (soft Small Medium Large
1st filesystem 1st filesystem bundle) *
descriptor descriptor
Spectrum Scale SW PID 5725-Q01
2x 128 TB to 512TB of Flash HW List Price ($600/TB raw) $76,800 $153,600 $307,200
Clients own x86 servers (Power support in Oct 2016)
Example Net Price (75%
1 year warranty (3 year warranty option in Aug 2016) $19,200 $38,400 $76,800
discount)
5 days (40 hours) of Lab services for configuration and tuning
Example SW $/GB raw $0.15 $0.15 0.15
Example SW $/GB if
18 IBM and Business Partner Internal Use Only $0.30 $0.30 IBM$0.30
Systems
mirrored
 According the Analysts, IBM is well positioned; however competitors
are moving into this segment as well
IBM offers the right flash for the This product completes the
Example This kind of product at this price point IBM all-flash storage portfolio
with support and services, is a game right workload for the right reason
quotes from - Mark Peters (ESG) for all workloads.
changer. - Eric Burgener (IDC)
Analysts about - Rich Fechera (Forrester )
DeepFlash 150

FlashBlade from PureStorage FlashScale from DDN

Still in beta Supposed to start shipping in August
Estimated $3 / GB (raw) A part of the SFA product line; fundamentally a block
No HDFS support storage offering that tries to be everything to
No policy based tiering everyone (Hyperconverged, support SSD, NVMe,
Not field proven as Spectrum Scale with performance nodes, capacity nodes )
4000+ customers Supposedly, can also run Spectrum Scale, but unclear
what level of software, and if there are any
Isilon Nitro from EMC optimization or tuning?
will come sometime in 2017 Intel solution for Lustre running on any Flash or
DSSD from EMC attaching to Spectrum SSD
Scale Open source software with limited function
Very expensive and low density 20 TB per U No single vendor support: Intel sells service and
High theoretical performance, but in real life support for the software only; hardware vendor sells
what do you get? What server and software Hardware support
19 can drive it? Little or no tuning for flash performance
IBM and Business Partner Internal Use Only IBM Systems
 Establish References with Top Use Cases
Target 4000+ existing Spectrum Scale clients and expand footprint

Data platform for Analytics High Bandwidth Data Tier Burst Buffer
Example: Hadoop, Spark, SAP Hana Example: Digital Media, Life Science Example: R/W buffer for HPC

Unified data repository, Move entire working data set to flash, high
Large dedicated tier to speed up
support multiple analytics availability through Spectrum Scale
writes & especially reads

Key Advantage: Key Advantage:

Faster time to insights
Load and off load data to and from
memory faster
Shared data platform for multiple
instance and forms of analytics
Key Advantage:
High bandwidth, low latency data tier Manage burst read/write patterns
common to HPC applications
Complete set of enterprise storage
Speed up MPI and check-pointing
services and enterprise availability

20 IBM and Business Partner Internal Use Only IBM Systems

 Performance

21 IBM and Business Partner Internal Use Only IBM Systems | 21

 Performance
Testing
configuration

with

Spectrum
Scale
IBM DeepFlash 150 IBM DeepFlash 150

22 IBM and Business Partner Internal Use Only IBM Systems

 IBM DeepFlash 150 Sizing Guide

23 IBM and Business Partner Internal Use Only IBM Systems

 IBM Spectrum Scale +DF150, Elastic Storage Server Performance

DeepFlash 150
490 Raw TB
DeepFlash 150 two 3Us, fully
20 245 Raw TB populated

GB/ two 3Us,

populated
s DeepFlash 150
123 Raw TB
Two 3Us,
populated

|
24 IBM and Business Partner Internal Use Only IBM Systems 24
 IBM DeepFlash Offering - What it is, What it is not?

DeepFlash 150 itself is JBOF (Just Bunch It is not a low cost FlashSystem; no
of Flash) densely packaged with SAS RAID function, no FC / iSCSI interface
interface
DeepFlas It does not work with Spectrum
h the It has lower write endurance, sub 1 ms Virtualize or, at the moment, does not
latency and high throughput 12GB/sec work with Cleversafe
Product
bandwidth per enclosure It is not optimized for OLTP, write
Together with Spectrum Scale, this (update) intensive workloads or
offering is targeted for unstructured workloads that require microseconds
data workload latency
Soft bundled Scale and DeepFlash 150 It is not a hard bundle offering of
DeepFlas deployed on customer servers software and hardware
h with Provide high availability and higher It is not a pre-integrated IBM solution
Spectru performance by mirroring data and currently does not offer software
m Scale RAID (GNR) capability
Spectrum Scale provides advanced
Spectrum Scale does not provide Real-
the storage functions, including encryption
time inline compression; it provides post
offering and compression
process compression, which would likely
have a material performance impact

25 IBM and Business Partner Internal Use Only IBM Systems

 The Right Flash for the Right
Workload IBM DeepFlashTM
IBM FlashSystem V9000 With Spectrum Scale
IBM FlashSystem
Storwize V7000F
A9000 and A9000R IBM DS8888
Storwize V5000F

Environments Virtual Storage Infrastructure Grid Scale Cloud Storage Big Data Storage Business Critical Storage

Heterogeneous Enterprise- Cloud-optimized (QOS, Multi- Multi-protocol support z OS Support

class Data Services
Dynamic Data Migration
Multi-Vendor Management
Data Reduction
Key Multi-site active-active
Attributes
Tenancy)
Predictable High Performance
with Data Reduction
Technologies (including
dedupelication)
Ease-of-management
Policy-driven tiering
Single namespace
Data ocean
High-performance file storage
High Performance
Highest Availability
z/OS (GDPS)
Power HA
Power i HA
Three-site/Four-site
Six 9s Reliability
Enterprise Scalability

Distributed block workloads Large-scale distributed block Distributed file/object High-availability

SQL Server VDI Hadoop Low RTO applications
MySQL Hybrid cloud Media Streaming High-performance OLTP
Traditional IT applications CSPs (Mixed workloads, SAS Real time analytics
Typical Multi-tenancy) Spark High-performance data
Workloads SAP HPC warehouse
VMware Content Collaboration
Exchange High-performance backup
target

26 IBM and Business Partner Internal Use Only IBM Systems

 The building blocks of IBM DeepFlash Elastic Storage Server TM

Spectrum Scale High

Highperformance
performanceParallel
Parallel File
File System
System
Seamless
Seamlessscaling
scalingofofperformance
performance & capacity
capacity

Spectrum Scale RAID 2 and3 3parity

2 and paritydistributed
distributedRAID
RAID
DiskHospital
Disk Hospitalfor
fordisk
diskmanagement
management
End2EndData
End2End DataChecksum
Checksum Tracking
Tracking

256TBofofFlash
256TB Flashinin3U
3U
DeepFlash 150
Cost Effective High Density Flash
Cost Effective High Density Flash

27 IBM and Business Partner Internal Use Only IBM Systems | 27

 Introducing IBM DeepFlash Elastic Storage Server TM

8X faster response time, 8X lower latency compared to HDD version*

ESS GF1 ESS GF2

DeepFlash
JBOF

Spectrum Scale
I/O server

DeepFlash
JBOF

1 Flash Enclosures, 7U
2 Enclosures, 10u
180TB of usable Flash; 360 TB of usable Flash
Max Read 13.6 GB/sec;
Max Read 26.6 GB/sec;
Max Write 9.3 GB/sec Max Write 16.6 GB/sec
*based on SPEC SFS results

28 IBM and Business Partner Internal Use Only IBM Systems | 28

IBM DeepFlash Elastic Storage Server Models

ESS GF1 ESS GF2

256 TB physical, 180 TB usable 512 TB physical, 360 TB usable

Spectrum Scale RAID erasure-encoding data protection Spectrum Scale RAID erasure-encoding data protection

13.6 GB/s large-block reads 26.6 GB/s large-block reads

9.3 GB/s large-block writes 16.6 GB/s large-block writes

Sub millisecond latency Sub millisecond latency

Spectrum Scale software + two Power servers Spectrum Scale software + two Power servers
+ one 256 TB DeepFlash 150 enclosure + two 256 TB DeepFlash 150 enclosures

Ultra-low power: 2W/TB Ultra-low power: 2W/TB

Dynamic capacity expansion (in 2017). Dynamic capacity expansion (in 2017).
Capacity upgrades without downtime, data redistributes Capacity upgrades without downtime, data redistributes

100/40 hours of lab services included for installation 100/40 hours of lab services included for installation

Linearly scalable across multiple Deep Flash ESSs

to 100s of GB/s throughput, many PBs of capacity, and 1,000s of IOPS.
Scalability is usually only budget limited.
29 IBM and Business Partner Internal Use Only IBM Systems
 IBM DeepFlash Elastic Storage Server is a building
TM

block
ESS is a building block that can start as a single array and grow into a
cluster, or can be added to an existing cluster:
an existing Spectrum Scale cluster (including non-ESS
configurations)
an existing ESS cluster (including HDD based models)

IBM COS
Spectrum Scale
DeepFlash ESS GL
BYO
ESS models
configurations

Single Name Space

Tier inactive data to a capacity optimized ESS or Object Storage on or off-premise based on a policy

30 IBM and Business Partner Internal Use Only IBM Systems |

 IBMs Deep Flash ESS and the Elastic Storage Server family
GS models use 2U 24x2.5 JBODs or SSDs, GL models use 4U 60x3.5 JBODs,
GF models use 3U 32 JBOF enclosures
Support drives: 1.8TB SAS, 400GB, 800GB, 1.6TB SSD 2.5; 2TB,4TB,6TB and 8TB NL-SAS 3.5
HDDs
Supported NICs: 10GbE, 40GbE Ethernet and EDR Infiniband

GF1 building GF2

block building
block

GS1 building GS2 building GS4 building GS6 building GL2 building GL4 building GL6 building
block block block block block block block

31 IBM and Business Partner Internal Use Only IBM Systems

Positioning of ESS models

Small File Cost

Throughput Large Block Cost
Random I/O optimized optimized
per rack sequential
performanc file/object
unit workloads throughput
e storage

DeepFl DeepFl
ash
ash HDD HDD HDD
ESS ESS based based based
8X
faster 2.8X ESS ESS ESS
response more
time

32 IBM and Business Partner Internal Use Only IBM Systems

 Advanced Research Computing (ARC) at Virginia Tech

4PBs of storage
managed

Business challenge: ARC systems support the full spectrum of high-performance computing needs in diverse research
areas such as astrophysics, bioinformatics, computational chemistry, engineering (aerospace, civil, electrical, mechanical
and other disciplines), geosciences, social sciences and several other areas. We currently offer over 4 petabytes of hard
disk storage running under IBM's Spectrum Scale. After more than doubling the number of cores in our compute engines,
we needed to add an all-flash tier to our filesystems to provide more IOPS and higher IO density (IOPS/GB of storage).
Solution: Deploy IBM DeepFlash 150 as a Flash Tier for IBM Spectrum Scale

The IBM DeepFlash 150 will meet our ongoing and future needs in a cost-effective way. -- Vijay Agarwala, Sr.
Director at Advanced Research Computing Group of Virginia Tech

|
33

33 IBM and Business Partner Internal Use Only IBM Systems

Standalone DF150 - Server and HBA
Support
IBM DeepFlash 150 is supported today on all Intel x86
Server must be compliant with one of the supported Operating
Systems
SAS HBA adapters: LSI 9300-16e or LSI 9300-8e
Intel Xeon 8-core E3/E5/E7 processor and at least 32GB of RAM

Supported OS:
Windows 2012R2 or higher November 2, 2016
RHEL 7.2 or higher
Announce support
Ubuntu14.04.1 or higher
by Power8

34 IBM and Business Partner Internal Use Only IBM Systems

 Standalone DeepFlash 150 online resources
IBM KnowledgeCenter Spectrum Scale with DeepFlash 150 Install, Config, RAS Guide:
http://ibm.co/2cHuUd6

IBM DeveloperWorks Using Spectrum Scale

with DeepFlash 150
http://ibm.co/2cCWbzk

IBM DeepFlash 150

KnowledgeCenter:
http://ibm.co/2cm2m7p

35 IBM and Business Partner Internal Use Only IBM Systems

Questions?