SPACRC

Is The Future Flash (Part 2)

In my last post I specifically used the term SSD’s when referring to NAND Flash Memory based disk drives as I wanted it to be generic and not be seen to have an EMC bias.

I did state that EMC uses the term Enterprise Flash Drive (EFD). I agree with Barry Burke that it is in the industry’s best interest to use this term for Enterprise Class Solid State Disks (SSD’s). Now some people would say I would, as I used to work for EMC, however those of you, who know me, know that I have my own opinions and will openly express and share them.

So why is it in the industry’s best interests?

From my experience of presenting and meeting with customers during 2008, especially the first half of, there was much misunderstanding about Flash memory and SSD’s.

When EMC announced its planned support for EFD’s on 14^th Jan 2008, there was the resulting FUD from some other storage array vendors, which resulted in misinformation circulating in the industry.

As an aside I remember this date well for a number of reasons.

1. Because it was exactly five months before my 40th birthday, not that I was worried about turning 40.
2. I actually meet with Barry Burke (The Storage Anarchist) in his office on that day, as I was over at EMC Headquarters in Hopkinton.
3. I was one of the few EMC field people who knew that EMC were going to announce Flash.

Before EMC introduced the STEC EFD, most people were aware of Flash memory based SSD’s, as they were starting to be available for laptops, all be it at a significant price relative to the laptop. And everyone (well perhaps everyone apart from your granny) was aware of Flash memory from the many modern gadgets that incorporate Flash memory such as Cameras, MP3 Players and USB memory sticks.

I know Barry did some excellent blogs back in early 2008 to help address some of this misinformation. I highlight three below.

http://thestorageanarchist.typepad.com/weblog/2008/01/0059-bold-fast.html

http://thestorageanarchist.typepad.com/weblog/2008/01/0061-swinging-f.html

http://thestorageanarchist.typepad.com/weblog/2008/03/0070-horton-hea.html

Obviously they contain Barry’s sense of humour (which I am not adverse to), and a slight EMC bias ! BUT they do contain the facts.

So whilst the term SSD is a good generic term, I believe it is appropriate to have a term that differentiates this class of SSD from the consumer class of SSD, which these days you can get from $100 upwards – depending on capacity.

So what defines an enterprise class Flash Drive?

An SSD based on Single Level Cell (SLC) Flash Memory?

(As an aside the term Single Layer Cell is also seems to be used.)

Well NOT any more based on STEC’s announcements on Tuesday.

Chris Mellor on The Register did an excellent summary yesterday (Wed 12^th) on STEC’s announcements, beating me to what I was going to include in my post. Not that I am trying to compete with Chris in any way!

 http://www.theregister.co.uk/2009/08/12/stec_range_extension/

I had expected STEC and/or others to announce an MLC based SSD any time soon, which they classed as enterprise.

To quote the STEC press release, based on its announcement at the Flash Memory Summit on Tuesday (the one I said I wanted to attend in part 1)

Until now, MLC NAND has not been considered a practical storage media for SSDs used in the Enterprise applications due to the rigorous endurance and reliability demands of those markets. However, STEC designs and implements high performance flash controller technology and proprietary advanced Flash management algorithms to render Enterprise‐grade performance and endurance that is superior to hard disk drives (HDD) and consumer‐grade MLC SSDs. In addition, utilizing sophisticated Flash management algorithms, STEC’s MLC‐based SSDs ensure constant Enterprise‐level performance across demanding workloads and data sizes. By achieving these critical thresholds, these drives are well‐suited for the requirements of demanding enterprise environments where SSDs will be deployed.

And as detailed in the press release and Chris’s article, STEC expect these MLC based Enterprise Class SSD’s to be shipping by Q1 2010, and have capacities of up to 800GB ! I assume the 800GB capacity is its raw capacity, and storage vendors will configure it to appear / offer lower capacity to provide ‘spare’ capacity to increase the SSD’s life expectancy.

So the use of SLC Flash memory cannot be used to define an Enterprise Class SSD, based on STEC’s announcement.

The problem with MLC Flash memory is that it can tolerate far less rewrites than SLC NAND. One can assume based on STEC’s statement that they believe (and I don’t doubt them), they can use MLC Flash and achieve good performance and reliability.

The STEC press release states the MLC based SSD is not quite as fast as their SLC counterparts, but offers not hard facts. My personal view is that performance of the STEC MLC based SSD is that it will be good enough for most customer requirements. More on that in part 3.

The press release appears to make no explicit statement about the STEC MLC based SSD reliability, compared to the SLC based SSD.

BUT it does talk about cost-performance advantages. Which to you and me means a lower cost per GB. I am sure EMC and others know what the price differential is compared to the SLC based SSD, but we will have to wait and see what the storage arrays vendors actually charge for them.

So within two years of EMC announcing their plans to support SSD’s, STEC have increased the raw capacity of its enterprise class SSD from 256GB to 800GB, and have introduced a lower class or tier of enterprise class SSD. So when I spoke to customers about 1TB+ SSD’s within a few years then I was not wrong, but it was not really a difficult thing to predict.

To get back to my original question, how do you define an Enterprise Class SSD (EFD).

My personal high level definition is:

1.    Is highly available. i.e. tolerates single component failure.

2.    Supports high read and write IOPS (small block I/O) and high bandwidth (MB/s), again both read and write.

3.    High MTBPR. i.e. Supports sustained I/O workload as defined in point 2, for several years before needing replacement or failing.

Now I know the above is vague, as I use the word 'high' with respect to availability, performance and longevity, and 'several' with respects to its longevity.

With respect to point 1, it needs to be as tolerate as possible to single component failures. This does not necessarily mean every single component within the SSD, but at least the Flash memory and DRAM chips (used as buffer/cache). No one in their right mind would advise to run an SSD unprotected in a storage array. RAID 5 should be considered the default, but as these SSD’s get bigger, in the same way SATA disks have, perhaps RAID 6 should be considered. The rebuild times for the SSD’s are quicker than HDD’s, but as the SSD’s get bigger the rebuild times will grow.

With respect to point 2, high is a vague term. MLC can offer good read performance even in consumer based SSD’s, and the performance of an SSD is down to its architecture and code, as the STEC press release states.

With respect to point 3, whilst point 1 covers its high availability, with a continual high I/O workload an SSD drive will eventually wear out. Enterprise class SSD’s can warn when they are nearing their end of life, and so can be proactively replaced, however customers will not want an engineer to continually replacing the SSD’s, as there is always risk with any replacement affecting a storage array. Five years seems to be a commonly quoted and acceptable minimum life expectancy, and in reality most current I/O workloads are not driving enterprise class SSD’s anywhere near their maximum capabilities.

So whilst it might be a bit of a marketing term, I do believe it is appropriate to use the term EFD, rather than SSD, and whilst I know I am guilty of using the term SSD throughout this post, the term EFD does help to different it from the consumer class SSD, and EFD should not be thought of as an EMC term, but a industry term, in the same way people refer to Hovers, rather than vacuum cleaners.

However as I have illustrated I believe it is hard to provide an exact definition for an EFD, especially given the fast moving development of new SSD’s based on MLC Flash memory. In the same way it is hard (impossible) to define enterprise and midrange storage.

My next post (Part 3) will talk about EFD performance and performance of EFD’s in storage arrays, and try to cover some of the other EFD vendors such as Samsung and Intel, as I might be accused on STEC bias. And no I don't have any STEC shares unfortunately.  

Is the Future Flash (Part 1) ?

This week (starting today – Tues 11^th Aug) there is a Flash Memory Summit in Santa Clara.

I would have loved to attend this, as there is a whole day’s track on Flash in Enterprise Storage today, and a morning of tutorials on SSD’s in Enterprise Storage on Wednesday morning, as well as some keynote sessions on Flash technology.

Unfortunately I could not justify to the wife, the cost of flights, hotel and summit fee, as it would have been at my expense. She would just think I wanted another holiday, as she thinks these conferences and summits are all talk and play.

Attending the Flash in Enterprise Storage track would have been really interesting, to see what everyone had to say about this rapidly developing area of enterprise storage. If you do attend I would love to hear a summary of what was presented and discussed.

Plenty has been written by analysts, consultants, media and vendors on Flash Drives / SSD’s since EMC’s announcement of its plans to support solid state drives (SSD’s) in DMX-4 back in Jan 2008 (EMC uses the term Flash drives, or Enterprise Flash drives rather than SSD’s).

Whilst at EMC, and since leaving EMC I have watched with great interest the ‘war of words’ between storage vendors who had announced support for SSD’s, and those that had not, and how this has played out as storage vendors have gradually announced their support for SSD’s over the last 20 months since EMC’s announcement. Most storage array vendors now support SSD’s in some form, or have announced their intention to do so.

So it is with interest, when I read a flyer on STEC’s website stating they have signed a $120 Million Supply Agreement for ZeusIOPS SSD’s (their enterprise class SLC based SSD) for 2H 2009, and are now forecasting total sales of these Flash drives to exceed $220 million in 2009 !

Now I don’t know what actual mix of SSD capacities they expect to sell in 2009, but revenue of $220 Million sounds a big number to me, which I assume translates into a fairly large number of SSD's. I honestly don’t know what Compellent, EMC, Fujitsu, Hitachi, HP, IBM and anyone else using STEC SSD’s, are paying for their drives, and it obviously depends on the capacity of the drive, which can now be up to 750GB, and the volume of drives they commit to. From some simple maths (UK English !), taking a guess at the cost of the SSD's, it must mean they are expecting to ship 10’s of thousands of SSD’s before the end of the year.

Here is where I wish I had followed my instincts at the beginning of this year when the STEC share price hit a low of $3.67. They had been recently trading around the $34-$35 mark, but have just dropped back to around $30-$31. Still over an 8 fold increase in value in 8 months!  I can only assume there are some very happy investors, whilst I am left kicking myself.

So based on STEC’s public statements there certainly now seems to be growing demand for SSD’s within the storage market. This shows how quickly the market can move from when EMC first announced Flash drives on 14^th Jan 2008, and started shipping at the end of March 2008 with the GA of 5773 code that supported Flash drives in DMX-4, later in the year support for Flash in Clariion (CX4), and this year with the GA of V-Max in April. And with the other storage array vendors now shipping SSD’s, it would now appear customers are now taking the plunge and purchasing some SSD’s for their storage arrays.

It is interesting that most storage vendors have followed EMC’s lead and initially chosen the STEC Zeus drive, though I am sure many storage vendors are in talks with the other SSD manufacturers to second source, given the other major name companies that are now shipping or developing SSD’s. The likes of Hitachi, Intel, Micron, Samsung and Seagate to name a few, have entered or are entering the Enterprise SSD market.

Interestingly companies such as EqualLogic introduced support for a 50GB Samsung SLC based SSD with a SATA interface in March of this year for its PS6000 series storage array, which from a quick check on Samsung’s website is a 2.5”, rather than 3.5” drive. For the record STEC do also offer a 2.5” version of their Zeus SSD drive. The Samsung SLC based SSD is a lower performing (lower IOPS / MB/s thoughput) drive than the STEC drive, and also a lower cost. So note, Enterprise Class SSD’s are not all equal. Something I will expand on in a follow on blog post.

Now I will openly state I am a fan of Flash. It is an amazing technology when you really think how much information (bits/bytes) is stored inside each Flash chip, and I believe, Flash memory has, and will continue to dramatically change storage arrays other the coming years. Note I say a fan of Flash memory and not Flash drives, though I do think Flash drives are great to.

My personal experience, from when I was at EMC, in 2008 and the first part of 2009 was that there was great customer interest in SSD’s, as I spent many hours presenting and talking with customers about SSD’s and Flash memory and the business impact and benefit SSD’s could have, however not many customers purchased SSD’s for DMX-4 during 2008. I often said to customers, it is not if you should implement SSD’s, it is when. I will also come back to this in a future blog post.

Since I started writing this post, I have found a number of interesting whitepapers on SSD’s that have been recently published, either directly written by vendors or by analysts sponsored by vendors, and rather than post one massive post, as this was turning into, I now plan break it up and delay publishing the rest of what I have already written, and expand on it further.

So stay tuned for parts 2 and 3, and perhaps even part 4 where I will enter into discussion/debate about how and where SSD’s should be used, the impact of SSD’s on storage arrays, and how I see future storage arrays and Flash evolving.

Why is Thin Provisioning Over 10 Years Late and What Impact Has This Had ?

I am sure you are thinking, what am I talking about, when I say:

Why is Thin Provisioning over 10 years late, and what impact has this had.

I will get to that in a minute.

As this is my first blog, I will just quickly introduce myself.

I am an enterprise storage specialist. In a later blog I will expand on what I mean by enterprise, which is a term that is probably overused and becoming harder to define with the blurring between mid-range and enterprise storage solutions.

I worked for EMC for over 9 years in the UK, with the last 4+ years as the Symmetrix presales specialist / product manager for UK&I, and 3 years prior to that at STK a technical presales role mainly selling mainframe tape and DASD solutions, and for 7 years prior to that at Lloyds Bank (a large UK retail bank) in the mainframe storage team.

I took a redundancy package from EMC back in March and have taken a few months off (mini career break) to work on a loft conversion with some builders on my house, and took the family to Orlando for a holiday in May and popped into EMC World at my expense to meet with many ex colleagues and see what EMC was publicly saying about V-Max and all their other products, and get the latest news and views in the IT and Storage space.

I have continued to follow the IT and storage industry since leaving EMC, and have read many blogs, though resisted the temptation to leave comments or blog until now. I do though have many thoughts to share, and will do so over the coming months. Thank you to those of you who have persuaded me to start blogging.

And I am now in job hunting mode, so any offers welcome ! - Sorry for the plug (advert).

Anyway back to the purpose of this blog.

What do I mean, when I say why is Thin Provisioning over 10 years late.

I assuming the reader understands the basics of Thin Provisioning. And I know HDS, EMC, IBM etc, use other terms to describe their Thin Provisioning implementations, and much has been written about Thin Provisioning over the last few years.

The major/classic enterprise storage vendors (EMC, HDS and IBM) have all only introduced Thin Provisioning capabilities for their enterprise storage solutions within the last year or two, with HDS been the first. I think is is fair to say that at the time of introduction none were production ready (for example lacking replication support), and it is only within the last year that their customers might have considered deploying Thin Provisioning in a production environment, using these vendors products.

I know 3PAR, NetApp and other vendors traditionally thought more of as mid tier players have had Thin Provisioning capability for slighlty longer. More about this in another blog.

Some of you with a mainframe background might remember the STK Iceberg or IBM RVA (RAMAC Virtual Array - OEM of the Iceberg) from the mid / late 90's. The Iceberg storage array provided Thin Provisioned storage (DASD) in an IBM mainframe OS/390 environment, as well as compression ! Though STK and IBM factored in compression when selling the Iceberg, selling it based on effective rather than raw capacity of the installed disks. And as IBM rightly described it, it was a virtual array.

As a customer I used the Iceberg (I was involved in the purchase of a total of approx 18 over a 2 year period at Lloyds). Its Thin Provisioning functionality significantly simplified the provisioning and management of the mainframe storage environment, and provided great flexibility that I will expand on sometime in a future blog.

So why was the Iceberg NOT a huge success and why did it NOT take over the world. The mid / late 90's were when EMC became dominant in the mainframe market, and whilst IBM OEM'd the Iceberg in June 1996, and gained a reasonable market share with it over the following 3 years, EMC became the No 1 player, and the STK/IBM OEM agreement came to an end in 1999/2000 when IBM introduced the Shark.

I could write much more on the Iceberg, the benefits it gave customers, and why I believe it did not continue to be a significant player in the 21st Century.

The question I am asking, is why did it take so long for EMC, HDS and IBM, who all played in the mainframe storage market in the 90's, and were fully aware of Iceberg, not see a need to develop Thin Provisioning functionality for their products back in the 90's or even the first few years of this century .

Was it because:

1) They did not understand the benefits Thin Provisioning gave customers, so did not see the need to introduce this type of functionality ?

2) They thought it meant they would sell less storage. Which in turn would mean less revenue ?

3) Large customers were not pushing for Thin Provisioning functionality, but other functionality, such as advanced remote replication ?

My personal view is that it was a mixture of all these factors and probably other factors, because even when I spoke with someone technical at EMC, who I respected back in Jan 2008, and said to them that EMC needs to provide Thin Provisioning support on Symmetrix for the IBM mainframe, they said why, and did not appreaciate the benefits it could deliver.

So what impact has the lack of thin provisioning over the last 10+ years had.

It has meant:

• Storage utilisation levels have typically remained poor at well under 50% for many years. Obviously this figure varies based on each customer environment though I am sure no one would argue utilisation levels have been poor.
• Customers have purchased far more raw storage than they would have needed if Thin Provisioning had existed. Which means capital expenditure was potentially far greater than it should have been.
• Significantly more floor space, power and cooling have been consumed, than would have been necessary. Meaning building of new data centres could have been delayed, and less CO2 would have been emitted into the atmosphere.

So the lack of Thin Provisioning has had a huge impact on capital expenditure and TCO, as well as the world's energy resources and CO2 emissions !

IF Thin Provisioning had been commonly available from even 2001, once Y2K was out of the way, it would have now become a defacto feature that customers would use without question, and have much experience of using, AND integration with Operating Systems, File System, Databases etc would have been developed, to enable Thin Provisioning to deliver its true potential.

However my experience is that the majority of customers who have purchased HDS and EMC arrays within the last year, are still only thinking about implementing Thin Provisioning. And when I have meet with, and presented to customers over the last couple of years there was a big relunctance, perhaps fear, of exploiting Thin Provisoning, and provisioning more virtual capacity than real capacity exists, for fear of running out of space, as well as concerns about performance.

However from my experience from the Iceberg product, and from storage vendors current products, provided they are sensibly implemented, with awareness of some considerations at implementation time, Thin Provisoning is a big win, both from ease of management, automatic data layout, performance optimisation, as well as the capacity/space savings with increased utilisation, which result in moneytary, power, floor space and environmental savings, and also less obvious operational benefits, such as been able to very quickly provision temporary capacity for development testing over a weekend.

I know HDS have tried to encourage customers to adopt Thin Provisioning with attractive pricing for the first few TB's, and EMC have just recently announced Thin Provisioning is FREE for all Symmetrix which supported it, which I believe is a good (if little late) strategic move to make Thin Provisoning de facto. This is something I pushed for since its GA when I was at EMC, and I know Martin (Storagebod) has been vocal on this.

My point though is that the slowness of HDS, EMC, IBM and others to deliver Thin Provisioning has had a big impact.

In the past generally it seems to be the smaller storage players that have driven true innovation, such as STK , and more recently 3PAR in this case with Thin Provisioning for Open Systems. I think/hope that might be changing as the big players listen more to customers, and a wider selection of customers.

For example EMC lead the enterprise storage market with the introduction of SSD/Flash drives in Jan 2008, though the ability to easily exploit them will not come until EMC delivers Fully Automated Storage Tiering (FAST). I will do another thought provoking blog on SSD/Flash drives in the near future.

I believe if storage vendors can deliver Compression and/or Deduplication, as well as features like EMC's planned FAST functionality, built on top of Thin Provisioning for primary storage, and deliver the same or better levels of availability and performance, this with really revolutionise storage over the next few years. I do believe Flash will have an important part to play in this to meet the performance demands of such a solution.

For now, all I ask is that ALL storage vendors continue to drive the development, integration and exploitation of Thin Provisioning to deliver business and environmental benefit.

I could write a whole book on the value and benefits of Thin Provisioning and why I believe the industry has been slow to develop and adopt it, but for now your thoughts and comments would be appreaciated.