Demystifying SSD, Mark Smith, S3

S3 INTRODUCTIONTrusted Advisor of Choice

Demystifying SSDFlash is not always a good thing (an independent view)

Mark Smith 21st May 2013

1 of 197

Trusted Advisor of Choice

Mark SmithTechnical Director


►Solid State Solutions (S3)› Leading independent storage integrator

› Big data and virtualisation

› Reduce risk and cost

› Flexible range of services

►Celebrating 25 Years› Privately owned and financed

› Customer focused, vendor independent

› Technically excellent, self sufficient

› ISO:9001 and ISO:27001 certified


►What makes S3 different?› Solution focused not product biased

› Technically excellent

› S3 Support

› Enduring relationships

►Numerous accolades› EMC Isilon Partner of the Year - Single Country EMEA 2013

› EMC New Partner of the Year 2012

› Nexsan EMEA Partner of the Year - 2008, 2009, 2010, 2011

› CommVault Trusted Advisor of the Year 2012 EMEA


►Managed Services› Offload day-day management tasks

» Servers, Backups, Storage

› Focus on your projects

› Predictable and controlled costs

►Cloud Services› Archive and preservation

› 100% integrity guarantee

› Email Archive

› Storage


►Support› 9/5 on everything we sell

› Support Advantage

› 24/7 uplift

› Close 70% calls with no vendor involvement

►Professional Services› Methodology

› Project Management

› Flexible offerings

› Highly referenceable


►Training


►Our Partners


►Over


Demystifying SSDFlash is not always a good thing (an independent view)

Mark Smith 21st May 2013

Flash Memory

► Flash Memory is a non volatile semiconductor originally designed for digital cameras

► They were not intended to be written to in the way most applications do!

► The two big challenges are:

› They wear out

› Read / Write imbalance

► The design maybe the same but the secret is in the manufacturing

Manufacturing, screening and quality control has a massive impact of the performance and reliability of flash memory

SSD Drives

► There are over 300 SSD/Flash vendors (59 didn’t exist before 1st Jan 2013)

► SSD is just the productisation of flash based storage

► Different types of flash :-

› SLC

› eMLC

› MLC

› TLC

Single Level CellEach memory cell can either be on or off.Most reliable type of SSD with a typical write endurance in the 100Ks of cycles.Performance – FastestCost - HighestBit Density – Lowest – 1Error Sensitivity - Low

Multi Level CellEach memory cell can support two levels (voltages) so more bits can be stored per cell with a typical write endurance in the 1Ks of cyclesPerformance – MiddleCost - LowBit Density – Middle – 2Error Sensitivity - Low

Enterprise Multi Level CellSimilar to MLC but with an enhance error correction to give much lower error rates. Typical write endurance in the 10Ks of cyclesPerformance – MiddleCost - MiddleBit Density – Middle – 2Error Sensitivity - Low

Triple Level CellSimilar to MLC in design but 3 bits of data per cell. Lower cost due to increased density. Typical write endurance in the 100s of cycles.Performance – MiddleCost - LowestBit Density – Highest – 3Error Sensitivity - High

Flash Failures

► Flash memory fails in a very different way to traditional mechanical disk

► The unrecoverable error rate for a mechanical hard drive remains pretty constant over the number of writes to a drive

► Errors are typically adjacent sectors (like a scratched CD)

► Flash unrecoverable errors increase exponentially with every TB that is written

► Overtime the write payload causes more and more errors to occur

Unlike traditional mechanical disk, flash drives are highly unbalanced, performing much slower in write situations than they do under read only loads. It takes much more time to erase and write a flash cell than it does to read it. Typically 10 times!

Flash Products

Acceleration of specific workloads

Features› Uses flash to accelerate SAS› Limited amount of flash› Relies on traditional disk for

capacity› Working set of data can be

small› Flash cache misses can

cause performance issues› Limited enterprise features

(replication, snaps etc)

Flash products typically fall into four categories:

PCIe Cards

Acceleration of single applications

Features› The flash based I/O

available› No HA features (mirroring,

HA, etc)› Low capacity so specific

application› targeting required› Not shared across hosts› Most expensive (cost per

GB)

Flash Appliances

Acceleration of specific applications

Features› High performance boost for

existing storage › Can compromise data

integrity› Effects DR planning› Most useful for read biased

applications› Expensive (cost per GB) but

can accelerate a large storage estate

General purpose Tier 0/1 storage

Features› High capacity – all flash› Typically has enterprise

class features (replication, snap)

› Not the fastest › Scales out to the largest

flash foot print of any category

› High performance for all workloads

Hybrid Arrays Flash Arrays

Flash and Applications

Flash is fantastic, its fast, its funky so would you like to buy some please?

► Do you have an application that would benefit from flash?

► Do you know your application’s performance footprint?

► How does your application access data?

Read/Write Ratio?

Working set size?

Latency Sensitive?

Persistent or Volatile?

The “Write Cliff”

So what is the write cliff and should you be concerned about it?

► The SSD write cliff is the effect where SSD write performance drops off after all the free flash memory pages in an SSD have been initially written to and the device cannot provide enough free pages to keep up with subsequent write requests.

► Each new write request then requires the SSD locate a block that can be erased for the new data.

► If a block that needs to be erased contains active data, the active data must be written to a new location to free up the block to be erased.

► This process of copying valid data from one block to a new block, called ‘write amplification’ increases SSD wear and is the primary cause of the write cliff.

Wear Levelling

Why is wear levelling important?

► Wear levelling is fundamental to the endurance of the flash drive

► SSD drives now contain wear levelling algorithms to improve longevity

► Flash array manufacturers typically add their own software wear levelling and smart data placement routines to further improve flash drive service life

► Advanced wear levelling gives an eMLC drive a 9 year lifespan typically**

› No Wear Levelling

› Dynamic Wear Levelling

› Static Wear Levelling

No Wear LevellingEach flash block is permanently mapped to a O/S logical block. This means that every block previously written too must be read, erased and written. Highly written locations wear out quickly while others could be left unused. Blocks quickly reach their end of life and could render the drive unusable

Static Wear LevellingStatic wear levelling works the same as dynamic wear levelling except the static blocks that do not change are periodically moved so that these low usage cells are able to be used by other data. This rotational effect enables an SSD to operate until most of the blocks are near their end of life

Dynamic Wear Levelling Each time a block of data is re-written to the Flash memory it is written to a new location. However, blocks that never get replacement data sit with no additional wear on the Flash memory. The drive may last longer than one with no wear levelling, but there is still an uneven wear pattern

** Pure Storage rate their SSD drives for 9 years due to advance software wear levelling

The Secret Sauce

So let’s talk about Flash Arrays …..

► Most new SSD/Flash arrays regardless of whether they are All Flash or Hybrid have a software layer. It’s no secret that most of the intellectual property is in the software.

► Commodity hardware – is it a good or bad thing?

› Less investment in hardware means more to investment in the software

› Easier supply chain with less chance of manufacturing delays

› Are they a software company looking to productise and sell up?

› Technical impact of commodity hardware, will I be hardware constrained?

► Deduplication technology – faster, slower?

› If dedupe was designed in from the start then should be faster with better drive wear characteristics and higher capacity

› Dedupe as an after thought just means wasted I/O

► Replication, snapshots and all that stuff

› Just like deduplication, designed in is great, thrown in as an after thought is bad

VDI

Lets look at VDI for an example Flash workload….

► 1000 users with an average of 25 IOPs per user

› 25,000 IOPs (+5,000 IOPs for tertiary workload)

› 2.8TB of storage space for OS and View components

Traditional Disk

► 166 data drives, plus parity and hot spares totals 194 drives!

› 20 x 8+1 Raid groups

› 100TB of available capacity (only need 3 TB)

› 20 x 1500 IOP Islands of performance

› 48U of rack space

› Power, cooling and purchase/maintenance costs

1000 users running Windows 7 Pro with local anti-virus (75% linked clones)

VDI

So how did we do it …..?

► All Flash commodity array with 18 x 200GB eMLC drives

› FC connectivity (for no other reason than they have Brocade SAN in place)

› 3.1TB usable capacity

› 90K IOPs maximum performance across two storage pools

› 2U of rack space

› 335W peak power consumption

The savings

► The solution showed a compelling ROI based on traditional disk

› Almost 4 times the required performance

› Much lower latency ~1millisecond

› 95% less rack space

› 94% less power

› EUE was excellent

VDI 2 Years on..

So how are things working today …..?

► Project was completed in March 2011

► Grown to 1300 desktops to date

► Write Cliff has not become an issue

► Reaching its maximum performance potential

► Good end user experience testing and user satisfaction

► VDI is an island of storage

► Gold images and persistent data stored on enterprise SAN

► Did not need to deploy replication or storage based snapshots

Which technology do I choose?

Think about your applications …..

► Flash based storage is great but they come in lots of flavours

› PCIe Card, Flash Appliances, Hybrid Arrays and all Flash Arrays

› Do you really need 800,000 IOPs?

› Low latency is usually the real answer

› If you copy TBs of data look for devices that actively manage wear levelling

Think about your investment …..

► With so many new flash vendors how do you avoid buying a dead duck?

› References, case studies, UK presence

› Investment, funding and the goal of the company

› It’s easy to make claims but can they back them up?

› POC, conditional P.O, labs sessions

› VMUG, Talk to your peers!

1 Prisma Park, Berrington WayBasingstoke, Hampshire, RG24 8GT

0870 7776111

[email protected]

We are here to help

Contact

Demystifying SSD, Mark Smith, S3

Technology