Barry Whyte is a 'Master Inventor' working in the Systems & Technology Group based in IBM Hursley, UK. Barry primarly works on the IBM SAN Volume Controller virtualization system. Barry graduated from The University of Glasgow in 1996 with a B.Sc (Hons) in Computing Science. In his 12 years at IBM he has worked on the successful Serial Storage Architecture (SSA) range of products and the follow on Fibre-channel products used in the IBM DS8000 range. Barry joined the SVC development team soon after its inception and has held many positions before taking on his current role as SVC performance architect. Outside of work, Barry enjoys playing golf and all things to do with Rotary Engines.

Andy Monshaw interview at eWeek

Two posts in the one day - that's probably a record for me, but this one is really just a link. An interesting interview with IBM Storage General Manager, Andy Monshaw over at eWeek. Andy describes, amongst other things, some of the customer business benefits Storage Virtualization can bring, some of which we did not fully appreciated when we began working on SVC, nonetheless many of our customers were quick to point out how great things were. One in particular Andy makes reference to and Tony discusses [vendor lockin] more in his recent post.

HDS making a great case for SVC

Its been a while, and recently I've not had the time or inclination to comment about Hu's blogketing, but a few of us found his recent post amusing. (As we often do)

I guess Hu has a dilemma. On the one hand he's been telling us for a couple of years that Storage Virtualization in the array is the only way to do it, and you should buy USP, USP-V (or one of their clones). But now he has to make a case for the appliance too. I've always said that technically there shouldn't be an real difference, when done correctly. Having some internal storage with your virtualization controller can solve some problems and provide an almost 'stealth' route to sell virtualization. Overall however he makes a great case for SVC... But two major points stood out in his post, and they both came in the same sentence!

When it comes time to replace the USP VM, you can continue to depreciate the storage system that you bought later by attaching it to the next generation USP V or VM.

One MAJOR thing missing here is that he doesn't mention that replacing it will be disruptive. Just like it was migrating from USP or NSC-55 to USP-V or USP-VM.

Secondly, he tries to suggest you still need a big box too, you still need the USP-V - why? If the USP-VM could provide the performance needs of a true clustered appliance model, why would you want to buy a very expensive second layer of virtualization? Maybe its because the USP-VM has mediocre performance (but a low entry price) whereas USP-V has higher performance (and a much higher entry price). But you can't upgrade from one to the other. Go figure. As per discussion comments on a recent post of mine, most SVC users see that all you need is a low cost, reliable, reasonably performing RAID brick.

Anyway, as I've covered many times before, there is no point having a great solution to solve your storage controller upgrade paths (online data migration etc) when you simply move the upgrade problem to the level above.

Again, as I've hopefully drummed into you, SVC can be upgraded without disruption from the smallest to the largest system and we can completely replace the SVC node hardware with the latest and greatest platform technology without disruption either.

So, remind me again, which provides for your virtualization needs and solves your upgrade woes?

Why virtualize - part 1

Since the explosion of interest in Host based virtualization products like VMware, Zen and Microsofts Hyper-V, it seems like the IBM concept from the 70's mainframes has really hit home. I remember sitting in on a very early customer focus group when we are mid-development of SVC and the speaker was discussing how virtual memory is taken for granted today, you don't really think about your pagefile. The speaker went on to describe his vision for storage virtualization, the names are the same because the concept is the same. Abstraction. Let one part of the system think its using another part when really its something totally different. The speaker went on to explain that he felt Storage Virtualization would become the same, i.e. something you take for granted. So why would we want to do that? Well thats what I plan to outline over the next few (probably weeks) of posts - unlike Tony who manages to nicely week'ify his topics, I seem to manage a month'ify at best!

My initial thoughts are to cover the following ideas :

• The problem
• The solution
• The benefits
• The gotchas
• The things to look for in a solution
• Performance Considerations

This list is likely to change as I write, and get feedback but thats where I'm heading over the next few weeks - feel free to ask for any thoughts, topics or content and I'll do my best.

As an aside, Windows pagefiles... Most of your are probably using some flavour of Windows to read this page, if you are, where is your pagefile? Have you thought about it recently? Probably not, most likely you have left Windows to create it itself, manage it itself and then complain when your hard drive starts thrashing.

One of the great un-features of Windows is that is can't defrag 'unmovable' files i.e. the pagefile, the registry and your hibernate file if you have one. Ok so page defrag is a great util from what were SysInternals, which Microsoft bought up! (Says a lot!) Anyway, the best solution I've found is to always create a small partition that is just over double the size of your memory (by about 256MB). Once you've then installed Windows change the pagefile to reside on the small partition and create it equal to the size of your memory. Then there is enough space to defrag it (as even page defrag needs twice the space to enable a copy of the pagefile) and as long as you don't use the small parition for anything else it never gets fragmented. This I've found stops that pause for 30 seconds with massive HDD activity on Windows OS's. I've started to get pee'd off with my hibernation wake-up taking ages and I'm just about to create another small partition to try the same trick with my hibernate file!

Evolution of Storage

One of the nice things with working on SVC is actually helping customers to solve their day to day storage headaches. We all know and feel the pains caused by the ever increasing amount of storage media needed in our SANs. The management costs, from a people and resource point of view alone start to spiral out of control and adding yet another box that needs another set of skills to maintain or provision becomes a nightmare.

Thats where storage virtualization comes in, and in particular SVC. Over at www.information-age.com there is a great example of how the London Natural History Museum have used SVC and IBM Disk/Tape products to tame their SAN. The Natural Selection of Storage.

The solution was implemented with the help of IBM UK Business Partner, Tectrade and in the customer's own words has provided exactly what SVC can : Quotes from Gavin Malarky - senior infrastructure analyst at the museum.

"The ease of management that virtualisation gives us made our lives a lot easier – flexibility and expandability within a single management console,"

"We don’t talk about archiving data any more, we just move it from online to near-line"

"The benefit is that, if we want to add another 10TBs, we can get another IBM DS4100 or buy disks elsewhere, sit them behind the SVC and, at the front end, it is all the same – simply a storage resource to the server"

The IBM SVC product pages have a detailed case study covering the Museums solution, for anyone with similar headaches I'd recommend they have a read and contact IBM or their local IBM Business Partner to discuss their needs and see if SVC can help them in the same way.

... oh my!

Over at Burke HQ, I see that he's a little put out by a few questioning articles after EMCs 1Q broadcasts.

In his latest post titled "Lions and Tigers and Bears" I can only but say "oh my!"

Its unlike BarryB to tee himself up for such an obvious retort and to use his words not mine, let me explain.

While discussing "Thin Provisioning", or as EMC are calling it "Virtual Provisioning" - now a quick aside - don't be fooled, this naming is a deliberate attempt to try and use the 'virtual' buzz word. Even although SVC is a virtualization appliance, IBM decided not to use the word in its name, instead calling it the SAN Volume Controller ™ or SVC for short. We have also decided not to use the term "Thin Provisioning" for our own in-house feature, instead it will be known as Space Efficient Volumes/Vdisks (SEV) which, due to the design of SVC's layered stack, gives us Space Efficient Flash Copy (SEFC) for virtually free, simply create the target Vdisk as an SEV and there you go. Anyway I digress as always! BarryB talks about Thin Provisioning and why it won't be of any use to Mainframe (zSeries) attached hosts - therefore by his implications, only open systems are likely to use it...

"And just as with RAID 5, Virtual Provisioning won't be deployed for everything in the data center - particularly in the markets served by Symmetrix. Databases don't really benefit all that much from thin provisioning, as they like to consume all the storage allocated to them and manage its utilization themselves. Similarly, the mainframe storage market won't be using thin provisioning to improve utilization - in fact, mainframes have historically operated at storage utilization levels literally unheard of in the open systems market. And where ever possible, snapshots are already "thin", so there's not a lot of opportunity there either."

On many occasions I have questioned why a big monolothic box needs to support every flavour of disk drive, instead it would be better to have a big box that caters for big hosts or applications, a middle sized box that caters for big or medium hosts that need reasonable performance, and then a cheapy box that caters for cheap SATA hard drives. As for flash, the only real enterprise ready devices are those from STEC, who may be in trouble if Seagate get their way, but are off limits to the rest of use due to an exclusivity deal - nothing we can do there, but there are many fingers in many pies and all I can say is things will get interesting very soon now. Anyway the point here is why would you want one big expensive box that was a 'jack of all trades and a master of none' (certainly not a master of cache unfriendly workloads)...

BarryB finishes by asking...

"In fact, instead of worrying about whether or not thin provisioning is going to reduce capacity demand for Symmetrix, methinks we should be asking "what about those products that don't even offer thin provisioning, or huge SATA drives, or super-fast Flash drives, or more than 1024 disk drives in a single array?" Why should they be getting a pass on these blatant gaps in their products just because they're spending money on Israeli companies?

You know who I'm talking about.

So lets think about this, if for example you had an appliance that could front all storage types, provide you with online data migration between said storage types, let you manage copy services across them all, soon provide Space Efficient characteristics, natively support any SATA or flash device you decided you wanted, provide many thousands of disks behind a single management interface and integrate with all the 'Israeli' products you could imagine... why would you care that just one of your products that has its largest footprint as a Mainframe box didn't have all of those features, when according to Mr Burke, everything the Mainframe does well it does itself, and by his own admission won't need or use features like Thin Provisioning.

My question would be, where is EMC's appliance that can do the same, is it Invista? Surely if they had such an appliance that worked, they would be bashing IBM with it, as they try to do with their big 'jack of all' monolith. Best of all they wouldn't have to force customers to buy the big monolithic DMX just to get these functions or features, remember, today's buzz is engery efficiency, isn't that Symmetrix a lot of data center footprint and power to run a dozen STEC drives flat out or even to keep a few hundred SATA drives spinning... Do yourself a favour, don't buy the FUD.

Update 24th April

As a bit of an aside, I'm told we have shipped more than 12,000 SVC nodes, at 4,000 customers - taking EMC's Invista claims of 200 installs, thats 20x the number of installs, and most of the SVC installs are in production use - anybody out there actually using Invista in production?

The maths must be contagious

So I'm still on holiday, but while getting myself prepared for the final stage of the re-decoration efforts at home I was browsing while having my morning intake of caffeine and came across an interesting viewpoint reported at Blocks and Files claiming that HDS are the #1 in storage virtualization...

I did a little digging and found the report on TheInfoPro (select storage on the right and it wants some details before it lets you view it). Now its difficult to actually see for real, but on the slide on page 3 it looks very much like IBM is ahead of HDS if you include in pilot projects. So have HDS been sharing their mathematical models with this research group too?!

A couple of things worth noting :

• The numbers here are comparing a stand alone virtualization device (SVC) with a storage controller that happens to have virtualization capabilities (USP) - nit picking maybe but think of it this way, there is a continual need for storage purchases which people have budgets for and is a standard IT asset. This means that its much easier to sell a USP by first going in as a storage controller, that happens to provide virtualization capabilities. SVC's numbers are there on their own merit.
• I found the link, and I was correct to remember a quote from HDS themselves that claim at any one point in time, about 10% of the USP in production are actually virtualizing external storage - mostly they are used for dracula type functions to pull in external data, or archive off data. So should we reduce this number quoted by TheInfoPro by 10%
• What happens when it comes to replacing the USP in a few years time? So you want to upgrade to their USP-V+, you need an external tool to migrate the data out, which is likely to be disruptive - doesn't that undermine all the benefits and the killer app of virtualizaton? Quite often we've found people will use SVC to solve this problem, and a lot of them leave SVC in there and give up on the USP virtualization, finding SVC much more flexible in the longer term. Thats why we support the Tagmastore and USP family as storage systems under SVC.

Anyway, statistics can tell you anything you want, and all vendors do it, but it would be nice to compare apples with apples, people who bought the USP primarily for its virtualization capabilities and not as a storage brick - however such data is subjective, difficult to obtain and probably not worth the effort. If you are one of those customers, be sure to contact your IBM sales rep when it comes to replacing your USP, you may find that SVC can solve the problems without resorting to days or weeks of downtime while you replace it.

Update: 21/4/08 : 10am I can't find the 10% reference, but I did track down a claim of only 50% of the USP out there have been sold with virtualization licenses. Case closed.

"IBM claims to have sold 10,000 of the SAN Volume Controller storage virtualization that it launched in 2003. Hitachi's virtualization software runs on the USP disk array that was launched in 2004, and the Japanese company claims that over half of the 6,000 USPs and diskless variants have shipped with a license for that virtualization software" CBR Online

Update: 24/4/08 : 1am A colleague found the link (thanks Chris), here HDS quote the 10 percent of virtualization use. Amended text in main body.

Get the painters in

After a few hectic months, I'd booked a week or so off work so that we could take the kids up to see their grand-parents. I originally hale from Glasgow and my parents still live there. My sister works in Edinburgh as a special-needs teacher and recently had a baby girl, so the plan was during the (rather late) Easter school holidays, to make the 400 mile trip back home to catch up and meet my niece. The best laid plans and all that. My son was off school the week before the holidays with, what I can only describe as full blown flu. The next week my wife caught it, so we postponed the trip a week, and yes a few days later our daughter came down with it. I felt like I was getting it and guzzled some vitamin C and other such remedies and managed to escape with a few days of rusty throat. Anyway the long and the short of it is that we've had to put off the Scotland trip till the next school holidays and I am eventually taking my break as a 6 day weekend. The old car seat, baby clothes and my old PC (my Dad gets a good deal with my old machines as I upgrade - hi Dad if you are reading ) will have to wait a few more weeks before they find their new homes.

After our daughter (3) had drawn an almost perfect outline map of the Isle of Arran (including The Holy Isle) at the top landing of the stairs, we'd been talking about re-decorating. So this weekend we've had the painters in - a.k.a me and the Mrs!

So far I've replaced the carpet at the bottom of the stairs with wood flooring, blending that into the living room. Double coated the entire hall space, ending up with that nice "spray-spotted" paint-roller look on my hands and arms. Meanwhile last week we sold our old living room aquarium (20 gallon). The wall behind which were still the previous wall colour (the thing was just too heavy to move when full). So I dug out the remainder of the wall paint and covered it up. Going over some other patches where out daughter had shown her artistic talents too . The next morning my wife pointed out that she thought the colour we had in the living room was a bit dark, and "look how its changed since we painted it" - there was now several very much lighter patches where I've covered up the artwork! So the living room has a new "light stone" covering (Dulux Once) at least, and it really is one-coat stuff.

Oh yeah and on Thursday while the Sun was shining I made it out and hacked down the Japanese Rowan (thorny bugger) and next doors ever encroaching hedges.

These programmers hands are shot! Sitting behind a desk or rack all day really is an easy life when it comes to hard physical work! I'm beat!

Now I just have all that white gloss woodwork on 10 doors and doorways to look forward too!

Meanwhile back to the reality (of work) the Diligent acquisition has been announced, like Tony much of what goes on in my daily work I am not at liberty to talk about here until such time as its made public, and over the next few months I will make time to comment on whats coming, a lot of which will explain why I've been so quiet recently. Anyway, deduplication has been of interest for a while, and as always SVC is perfectly placed to integrate with the latest and greatest functions and products, yet again proving the fundamental value of an in-band appliance. I'm off to rub moisturiser into my hands, and could do with a good old massage for my legs and back! A few more days of holiday left...

SVC Cache Partitioning Redpaper Draft Available

I keep making excuses for not posting much recently, but I really have been very busy. One of the things I started what seems a long time ago now was a paper describing why, how and what we implemented back in the November 2007 release of SVC - with respect to Cache Partitioning.

Since no SAN in perfect, and you can't always assume that a storage controller is going to behave nicely with its peers, or at least not hit hardware or other such failures, we decided it was worthwhile setting some limits on how far a single bad controller could influence the overall environment. Think of this as a dynamic and automatic throttling mechanism when your host systems start to ask to much from the physical spindles.

In general cases, it should make no difference to overall performance. It simply starts destage operations from the cache a little earlier when there is a lot of write data for a single partition - (managed disk group) - the more worthwhile cache resources (i.e. read data) is not partitioned / limited.

Anyway I'd had a few questions from customers and account teams and thought it was worthwhile spending some time writing up why and what we did, plus a bit of how it works internally. (Without giving too much away!)

PS. As I understand it, the lack of cache partitioning (despite all Hu's shouting about global caching) is one of the main reasons HDS don't recommend you use any of that global cache for external storage under a USP / USP-V.

http://www.redbooks.ibm.com/redpieces/abstracts/redp4426.html

This is still in draft release, so any comments, or further questions I haven't addressed, let me know and I'll roll them in.

Still here

Where does the time go. Just a quick update to say that I am still here, been working far too many hours to get new stuff benchmarked, new projects underway, infrastructure upgrades done and setup some nice new bits of kit that arrived recently.

I had to disable all comments to my blog as I was hit by spam supposedly from community.webshots - so if you try and post from there - sorry we gottcha! The site has been upgraded and some better spam protection installed (so I'm told) so hopefully I wont have another weekend with over 1000 spam emails in my Notes!

Anyway the recent Q&A post has been re-opened, so fire away if you want to ask anything, and I've got a couple of draft updates in the writing - interesting to see Dell's recent 'SSD' comments, user feedback and so on - I certainly wouldn't buy a laptop with only an SSD and a paging operating system, but more of that to come. For now, back to checking the testcases, entertaining our customers and preparing the next release.

A break in the clouds

Another couple of weeks fly by, and far too many evenings have been spent on conference calls and the like - such is life when many of your close colleagues and executive team are six to eight hours behind you. Anyway, thats my excuse for not having posted much recently. I've been following the usual suspects blogs however, and I'm glad to see that there is a 'lull in the storm' or maybe thats a break in the clouds and our friends at EMC have stopped the IBM bashing for a couple of days, only to pick up the big stick and have a go at HDS this week - [the storage anarchist] | [storagezilla]

Anyway, I shall retract my head again before the bashing comes my way again

Actually, let me repay the favour for once, not making fun of out of context 'tape' statements, but what could be seen as nothing more than lies. I was reading up and following my pet topic of solid state storage for the enterprise and found an article over on TechTarget [Solid state] and was amazed at the (possibly misquoted statement) from Bob Wambach, senior director of Symmetrix product marketing for EMC.

EMC has built a series of flash drives that offer faster read/write performance, high reliability and data integrity, he said.

Now, correct me if I am wrong, but has EMC bought up STEC Inc without anyone knowing? The implications of this statement to me are that EMC are claiming they have themselves built the drives. Taking credit where its not due.

It will be interesting to see how these drives take to DMX, or customers for that matter, and how much of the 25us flash latency is diluted by the using system, but as the article states this is an order of magnitude better 'device' response time than standard HDD. However at 30x the price (presumably thats 30x the price of a complete DMX + HDD vs DMX + SSD) given that the raw devices are much more than that today.

Anyway, as Clod Barerra is quote to state, IBM Research Almaden are looking at various technologies. I recently saw a presentation by them on Phase Change Memory (PCM) which is still at the research stage, but looks like an interesting technology - without the need for the same levels of write endurance and read disturb protection. And as Andy Monshaw (General Manager IBM Storage) recently stated, IBM are seriously investigating an aggressive approach to make the most suitable use of these new technologies, and while all vendors may today have to make do with placing OEM or organic SSD's in their existing product lines, it's what comes next that is really interesting.

Do our customers want it?

I'm glad that BarryB is finding things amusing, while at least there is some humour there, I find it amusing how much time he spends advertising IBM products. Out of his last 6 posts 3 have been about IBM, in particular DS6000 and DS8000. I think we get your message BarryB, you *think* these products are dead. We know they are not, its against IBM's policy to pre-announce actual features otherwise I could tell you of all the things that are coming this year, but then I'd be shot.

I wanted to pick up on a couple of point that were extrapolated by BarryB. First of all I understand this interview to have happened several months ago, with a few recent questions tagged on after our announcements earlier this month. That said, I myself don't see the link between SSD and Tape... Well I think I can see what was being suggested, there is nothing better than proper HSM, i.e. keeping the Tier1 data to minimum and moving older stale data to tape or MAID type products. However, this may have not come across very clearly in the small quotation extracted.

The key point about features and functions is what business need is there for them. Are your customers asking for them, or are you simply following the crowd because vendor X has done it. That said, most vendors do follow the crowd, as usually there is a high demand for feature X or function Y. Again, I'm assuming, but I guess what Charlie was saying is that he himself hadn't heard of customers asking for certain functions. For example, the enhancements to copy service functions in DS8000 and SVC recently were prioritised over DS8000 direct SATA attachment for example (FATA has been available for several years). For example, once you have SATA you need RAID-6 so these two things go hand in hand. The business case for the enhanced copy services was greater than any requests for other functions, so these were prioritised over other things. So just because EMC has feature Z doesn't mean we have to - unless a large percentage of our customer base is requesting them.

As for Solid State Disks, well, I think you all know my stance on them, if you've been following this blog for some time you will know I played with some last year and wrote up my thoughts, but as for what IBM itself is looking at I'll leave that to Storage VP Andy Monshaw in an interview posted yesterday on Information Week. The key thing is that its not just about bunging a few of these in an existing box and leaving it there, its the whole she-bang; controller, software stack and application that needs integrating... Over at RupturedMonkey Nigel is asking if EMC will go down in history as a game changer on this one, I'd contest this and suggest that accolade will go to the first vendor(s) to truly offer up the entire potential that these new breed of disks can provide. That means really eliminating the wear out, using the cheapest and highest density Flash devices while maintaining reliability. Most important of all is the integration of the entire stack from the applications down to the raw zeros and ones... Can a specialist storage only (sorry Information Infrastructure) company do that on their own... no... the first will probably be one of Tony's one stop IT supermarkets...

More Q&A

Been struggling to find the time over the last few weeks to pay you all as much attention as I should Two weeks ago I had a flying visit to Austin, Texas for an internal Summit. In attendance were Distinguished Engineers, Senior Technical folks and other specialists in the topic at hand, covering IBM's Systems and Technology Group (STG - which storage is a part of), Software Group (SWG) and Research. It was good to see and hear everyone's perspectives on the topic and I guess one of the major advantages of being part of a 'supermarket' IT company (as Tony would call us) against being a 'speciality shop' which deals in just one of the cornerstones of todays IT infrastructure. I was there to present on a new project proposal we are working on and some general storage strategy work we've been looking at. It was great to get the overall perspective from the whole corporation and it made me think about a couple of things we hadn't previously considered.

Meanwhile back at the office we are busy getting the next set of SVC features, functions and enhancements ready for System Test which should be out in the field later this year. As always there's more we could be doing than we can do, and everyone is flat out keeping the cogs turning.

I just upgraded my home PC and been tweaking, boosting and tuning to get the absolute stable maximum out of it - maybe thats why I've not been on here for a while... that new Radeon is too good to put Half-Life2 down... yes I know Nvidia are generally faster, but the price of the HD3850 was too good to be true and there is much to be said by brand loyalty (AMD and ATI here I'm afraid!)

Anyway, this one is over to you all, fire away with whatever questions you may have and would like to discuss and I'll see what I can do. Here is a 'starter for 10' that came in from a reader that wished to remain anonymous.

"If I go with a full virtual environment, servers and storage, are there any storage pitfalls I need to think about?"

So this one is one of those - it depends - answers. One major consideration is the actual requirements of all the host applications. So although products like VMware are great for Windows applications, where each application only uses a small amount of the CPU power available, they all need some form of storage performance to work at an optimal level. There is no substitute for spindle count, or at least raw I/O performance. You do need to still scale your storage environment to cater for the peaks in your application workload. While storage virtualization can allow you to squeeze that bit more out of what you have, the physical devices still have a limit, and if you go over that limit you do risk impacting all the hosts that share the same devices. the number one gotcha would be to try and squeeze a bucketful of I/O from a thimbleful of hardware. Planning, planning planning.

Hope that helped a little. Its a topic that probably deserves an entire redbook!

Here is the SMB data migration

I was pointed to an article on TechTarget by Tory Skyers where he vents his frustration / asks what aren't people making products that will solve their data migration issues.

Whether he's been looking at software or backup solutions, and missed the key killer app that storage virtualization... he asks :

If you sense some irritation on my part, you’re spot on. While it may not directly affect me, it most certainly does affect those who rely on my expertise. I’d like to be able to say "Hey, just get Product X from Vendor Y and it will take care of at least 60% of your migration woes."

My answer is "Hey, just get SVC from IBM and it will take care of 100% of your migration woes"

He talks of cost, and a base pair of SVC nodes - more than enough for most SMB's - won't set you back anywhere near the $50,000 he mentions.

Once virtualized behind SVC, all data migrations can be performed while maintaining online I/O access to the server applications. Just ask any of our reference customers, for example removing a 3 month phase out, phase in cycle of old to new storage, perform it during the day, over a couple of days and you can easily move from one controller to another - then extract the old controller. This is always seen as the killer app for the guys on the ground.

While Tory wants a real combined ILM, policy based data destroyer, SVC does only provide the data migration at a block level today, therefore its not going to solve all his needs, but its certainly going to make the data movement much much much less painful.

However, not only will you get the data-migration benefits, but you get the day to day storage virtualization benefits. Consolidation of your storage arrays, reduced complexity of management of all those small boxes, their mapping to hosts, and any copy services you use - all bundled under one single interface, provisioning methodology and if you need them a consistent set of copy services over all your storage. Not to mention the likely performance boosts SVC can provide when fronting entry and mid-range products.

Location does sometimes matter

One of my regular blog stops is over at Techworld where Chris Mellor discusses all things storage. I spotted an interesting post from Chris today comparing SVC to USP and coming up with the conclusion that the two are almost the same in what they can achieve, what they do and at a 40,000ft view, how they do it.

I came to a similar conclusion last year in my multi-part-work discussing Storage Virtualization (as we know it today).

I'd agree that the differences between what an appliance and what a virtualizing controller can do - when it comes to features - are pretty much the same. Assuming the controller can virtualize external storage (as is the case with USP) then you can apply the feature set across all your storage - unless of course you limit what people can do, due to either testing limitations or fundamental limits on the system - again as is the case with USP. One of the many benefits of virtualizing is the ability to extract old hardware while maintaining server access - again here USP has missed a trick - just how do you upgrade the USP to a USP-V... However these are all implementation details and not generic flaws of a virtualizing controller methodology.

However, where I do seriously disagree is with the 'on-switch' virtualization - as implemented by Invista. Here the very limited processing power, problems with global namespaces - fast meta-data updates across many blades - scale out issue across many blades - and so on do seriously limit the functions that can be performed on single instance blades. Hence why EMC need a separate Kashya blade to do replication, and why their Flash-copy is really just a simple split-mirror implementation.

For this reason, if Chris had left it at just a comparison between SVC and USP, I'd probably totally agree with him. Neither of these models technically limit the functionality and general 'comoditization' of the underlying storage devices. Thats one of the things we set out to do with SVC. Why have 10 controllers that all implement the same sort of copy services, advanced function, with 10 GUI interfaces and all subtly different. Turn them off, don't license them all, just use a single implementation - as in SVC or USP and then license what you need on the storage you want.

Chris ends by saying :

Unlike real estate the three most important things for a virtualising SAN storage product are not location, location and location. Let's forget about SAN location and concentrate on three far more important things: functionality, performance and price.

So if we exclude Invista or other switch based solutions I'd agree. In the open side of the world, SVC pretty much matches USP for functionality - or will do later this year when Thin Provisioning is available. Performance - since we both submit results to the Storage Performance Council (SPC) - not everyones favourite, but none the less, SVC can squeeze that bit more - but again both systems are as capable... However note that USP was with internal storage only - as I understand it there are quite hefty IOPs limits on the ports attaching downstream controllers - I'd love to see the comparative external attached storage benchmark. As for price, well SVC has a major edge on that one!

For those that missed my part work you can read via the following links :

• Part 1 : The Approach War
• Part 2 : Copy Services
• Part 3 : Performance
• Part 4 : Interoperability
• Part 5 : Upgrades

Can't wait for the excuses

So before Mr Burke, Chuck or the zilla get in there with the excuses let me have a guess at what's coming :

• They deliberately mis-configured it
• Its not a true comparison
• Its not a real life workload
• Its all about the number of drives
• ad infinitum...

I am of course talking about the latest batch of SPC-1 benchmarks, as discussed by Tony, and over at DrunkenData.

It's true, Chuck did give the go ahead for anyone to 'lash up' their EMC boxes and spin them at SPC, hats off to Net-app for going ahead and doing it, erm 'sponsoring' it. Now then, who's going to stick their DMX up for the offering and see what it can really do, and just how many of those SSD's you can expect to saturate

I really must not keep going on about SSD's like that. There are two uses as I see it :

1. Super IOPs uses - those applications that don't really need that much storage but have huge IOPs demands
2. Latency intolerance - those applications that don't care so much about IOPs but are killed by bad latency

Anyway, Texas Memory Systems provide the opposite, a play directly into value proposition number 1. Very very high IOPs rates, but at very low capacities. I take my hat off to, and I know Woody reads this, his companies new SPC-1 world record benchmark. Certainly adds a spanner to the works in the # of drives = SPC-1 number EMC equation. (I'm joking of course, its not the same thing).

I would question the $/GB of such a solution though, it does cost a premium over even the 30x SSD cost EMC have been discussing.

Anyway interesting times, who's going to be the first to 'sponsor' a DMX SPC-1 test... its got to be the best way to push EMC's hand to finally submit an official benchmark - unless they do, any finger waving or excuses about the Net-App results are just that, excuses. If they truly believe they can do better, then come on in.

Content revised 31st Jan to clarify / remove potential confusion.