Is EMC Developing Controller Based Virtualization?
I have heard from a very credible source (a recent EMC employee) that they are. It will be built into a future release of Enginuity and rolled out in the DMX. So what does this mean? Are they actually validating HDS's claim on where in the network virtualization should reside? Sounds like it.
So what would this mean for Invista? Have they finally realized that it doesn't work? Probably that customers just want something that can do replication AND all the other tricks that virtualization in a controller can give you. Why would a customer want to limit themselves to only being and to pool disk and migrate data around when they could have all of the great tools that EMC has developed over the years?
So here's my prediction: EMC will announce controller based virtualization some time this year, and either release it late this year or by June of 2009. They will phase out Invista completly or just use it as a data migration device in their services organization.
Anyone else have any thoughts on this?
March 5th, 2008 at 5:06 pm
If this is true then I will "literally fall out of my chair laughing. Not almost – I wiil actually hit the floor. LMAO"
But then after my ribs had ached for long enough I would catch my breath and think, "I dont blame them".
Personally Ive seen lots of customers who like it and use it well. So why wouldn’t EMC want a slice of the cake.
There will be those of us, probably me included, who will never let them live it down. But I dont think they will care as long as the pennies are rolling on off its back.
March 5th, 2008 at 11:55 pm
It may explain why even the storagezilla fails to bite at any of my attacks on Invista.As I covered in detail on my blog the SVC appliance model, and virtualizing storage controller model aren’t that far apart (when done correctly) – certainly from a feature, function point of view. HDS have for some reason added a lot of overhead to their implementation, some major limitations on what features they support on external storage and prefer to use it as a ‘dracula’ storage solution.If EMC do head down this route, then it will be interesting to see if they overcome these issues, and if the SVC they bought from us has helped them
March 6th, 2008 at 2:05 pm
I tend to lump the SVC and UVM implementations together for most things. The big thing I don’t like about the SVC is the whole vdisk thing. I don’t want a device that will own my LUNs forever. I want the ability to unvirtualize things if I so choose and I can’t do that very easily with the SVC.
March 6th, 2008 at 2:43 pm
you need the "svctask migratetoimage" command…….which takes the vdisk and turns it back into an image mode mdisk, which you can then unattach and present to a host again. All online of course……it doesn’t get much easier than that!
March 6th, 2008 at 5:50 pm
Am I the only one that thinks the silence from the EMC crowd over this is deafening!?
If EMC close the gap here, and possibly improve on the Hitachi implementation (entirely possible), then the USP will lose a key differentiator! We all know how much the folks at HDS like t point out EMCs lack of virtualisation in the DMX controller. I have personally been amazed at how long EMC have allowed this to go on.
Assume EMC do implement controller based virtualisation, although you can bet they wont call it virtualisation, the differentiators between the USP and the DMX will be stacked in the favour of the DMX.
May be this would provide a kick up the backside to the folks looking into fitting SSD into the USP and speed them along a little? After all, EMC cant be seen to have everything they have plus more.
PS. I hope you are right about this snig!
March 7th, 2008 at 9:35 am
Sounds like a rumor to me… In order to add Symm-based virtualization EMC would need to overcome 600TB usable capacity limit for DMX-4 which is impossible without significant cache amount increase which will drag additional batteries and so on. So, it can’t just be a software update.
March 7th, 2008 at 5:22 pm
What 600TB usable capacity limit. I have not heard of this and inquireing minds would like to know. How does this affect the "First Storage array to support 1PB" marketing speak?
March 7th, 2008 at 9:31 pm
The rumor is not accurate. There’s a lot of things to advance in the array, but virtualization, in the sense of HDS, is not one of them.What is definitely happening is to put a I/O splitter, similar to the recent announcement with regards to RecoverPoint and the CLARiiON. No longer will you need a splitter in the fabric (i.e. cisco and/or brocade). This will change the replication landscape and lower the cost point for SRDF-like functionality.
March 7th, 2008 at 10:09 pm
Re the 600TB usable capacity limit mentioned above. The spec sheet for the DMX-4 seems to suggest some limit around the 600TB mark –
This suggests an addressing limiation or something, close to the 600TB mark.
But hey, what I know about Symmetrix could be written on the back of a postage stamp.
Oh, and Stewey are you suggesting they may be working on virtualisation in the controller but implemented differently to HDS? And on what do you base your comment that the rumour is not accurate? At least snig half quoted a semi-reliable contact
March 7th, 2008 at 10:34 pm
Nigel, I’m suggesting that they are putting an I/O splitter in the array. I could see how I/O splitting could be perceived as virtualization because it uses virtual target/initiators, just like, ok exactly like, Invista. The implementation in the DMX is strictly to get ‘in the path’ of the data in order to write, in parallel, to a secondary target. It won’t be used for storage virtualization as it is with HDS. It’s CDP in the array.
March 10th, 2008 at 9:05 am
Morning Stewey,
Obviously I hadn’t read your previous comment correctly. That’s actually very interesting what you are suggesting, probably more interesting than just doing a USP style implementation.
With all the interesting stuff happening, and rumoured to be happening, with the Symm recently its starting to look like a real box of tricks. Almost makes me want to invest some time in learning it properly. Hmmmmmm…..
Nigel
March 10th, 2008 at 4:26 pm
Sounds to me a lot like the old Vicom solution. Also known as the IBM Piper tool.
http://www.vicom.com/
March 10th, 2008 at 5:53 pm
Thats interesting I thought originally that usable limit was there because 1TB drives weren’t available only 500GB I wonder if there is a real addressing problem or just lazy updating on a spec sheet. I didn’t see anything on powerlink and I can’t seem to find a datasheet that says anything different.As for write splitting I don’t see how that is any different then using an external storage volume as a USP BC S-VOL. It is still isolating the writes and dumping them somewhere else. I’m not an expert but I am assuming recoverpoint still needs to do an initial normalization so I don’t see anything really different here other than marketing speak. I guess you could say Hitachi doesn’t have a CDP solution integrated with external storage but I guess the underlying mechanisms are there it won’t be a huge issue to add feature parity to hitachi gear.
March 10th, 2008 at 6:20 pm
I guess it depends on how you define CDP. There are about 20 different definitions out there based on which vendor you are discussing it with. I guess we might find out EMC’s definition sooner or later.
I agree with you akro, I don’t see much of a difference right now between what EMC is talking about and what HDS is already doing.
March 11th, 2008 at 2:25 am
I have to wonder about virtualisation and anything new from EMC, et al. I have tried virtualisation with the USP and other systems such as 9585V and the LSI based systems like DS4800′s, etc. I put them into a seperate SLPR and carve up CLPR’s. But during heavy usage with the external systems, I get Write Pending issues. This in turn apparently causes issues with other CLPR’s and SLPR’s. I turned of any cache usage on the external LUN’s at the request of HDS but I still get some wierd and wild WP problems mostly during heavy backup times and I can live with that. The 9585V is the worst of the systems as it is 250GB SATA. The DS4800′s are 300 GB FC and they don’t seem to have the same amount of issues. That suggests that slow disk are not good to virtualise. Our HCAP uses 750 GB SATA so it better be a good solution as I don’t know what to do with it if we don’t keep HCAP.
Virtualisation is good but I wish it was better. I have always wanted to try the Cisco/EMC option just to see what it is like.
March 11th, 2008 at 7:01 am
arko:That’s an addressing limit. With current amount of cache Symmetrix can not address anything more than 600TB usable capacity. And EMC doesn’t like to admit it, they only do when faced with their own papers. Then they say smth like "we never tried to hide it". And "first 1PB array" is only in terms of raw capacity, not usable.
March 14th, 2008 at 7:39 pm
You all are using the term storage virtualization to mean one thing – the ability to consolidate various other storage behind a single controller/array. I would think this would have limited applicability, just mostly for customers who want to preserve their investment in the storage they have while consolidating. I think the virtualization within the single array, e.g. HP EVA is of more general purpose utility.
March 17th, 2008 at 8:42 am
2 billy bathgates: The whole post is about controller-based virtualization of external storage. Nevertheless, both EMC and Hitachi high-end storage have all kinds of "virtualization" stuff within array – from hypervolumes to thin provisioning. And EVA, even with its neat technologies, is not anywhere close to what real boxes can.Don’t read too much HP marketing
March 17th, 2008 at 8:54 pm
Well considering the EVA came out a long time before the current virtualization BUZZ I would say that their definition of Virtualization is fine. I always liken the EVA to LVM at the block level. LVM really is a disk virtualization solution for a host….
April 23rd, 2008 at 1:25 pm
well, the post is entitled ‘controller based virtualization’ – *without further clarification*. vIRTUALIZATION THESE DAYS CAN MEAN ANYthing the writer likes at the time, and they generally assume everyone somehow knows what they mean, which is a bit parochial, if you ask me.
I personally think the hypervolume style of disk management is old-style and a PITA to manage compared to several other arrays, and the EVA is the easiest and most sensible design of them all. However the higher-end arrays do have much more cache, backend etc, which may be required/justified in some circumstances, so in that case you are stuck with PITA management, for the time being.
April 24th, 2008 at 6:55 am
I think Billy likes his EVAs… LOL