EMC huffs, gives XtremIO a polish, shows off shiny data services

EMC's XtremIO all-flash array will benefit from new data services and models as the company spools up to take on Pure Storage, IBM and other flash array competitors. There is a new 5TB Starter X-Brick, lowering starter costs, and X-Brick clusters can scale out to six nodes with 12 active controllers. The Starter X-Brick can …

[Nate Amsden]

Petabyte scale

Let me do the math here..

6 x 20TB = 120TB raw

Assuming these new X-bricks are the same physical size as the previous units your looking at a full rack for 120TB of raw flash.

6:1 data reduction on 120TB means around 720TB "effective"

Maybe I missed a digit somewhere because I'm not seeing a petabyte here. (their data sheet tries to claim petabyte by saying if you take a lot of snapshots and stuff you can see a petabyte)

Seems like an improvement but still a pretty inefficient system. Would be curious why they limit themselves to only 20TB of raw flash per X-brick when that brick requires 6U of rack space. Seems like there is a pretty severe limitation in the software and/or hardware that they have. What if the workload is not de-dupe friendly? You have to buy a bunch more X-bricks because you need more raw capacity.

Can XtremIO be scaled out on the fly yet? I read or heard somewhere somewhat recently I think that you could not add more bricks to an XtremeIO system online, you had to buy the footprint up front(or perhaps take a big downtime for data migrations etc). Not sure if that was ever the case or if it was if it is fixed now or what. If it's not fixed yet I'm sure it'll be fixed at some point.

[AndrewDH]

Re: Petabyte scale

The available capacity before compression and De-Dupe is not 20 TB but 16.4TB per X-Brick.

This gives a protected capacity of 98.4 TB. Using your 6:1 compression + de-dupe ratio you end up with ~590TB which higher than the first release but then 6:1 is ambitious. Pure which has a somewhat more sophisticated De-Dupe scheme than X-tremIO is currently averaging 5.68:1, X-Termio is more likely to be closer to 4:1.

[Anonymous Coward]

Re: Petabyte scale

From memory each xbrick is made up of :-

2 x 1u Controller Nodes

1 x 2u Disk Enclosure - 25 x 2.5" disks

2 x 1u UPS Backup for controller cache

So that's 6u for 25 disks, but a facia is supplied to make it look like a single unit, however if you need more than a single xbrick then on top of that you also need a pair of infiniband switches for the backend cluster interconnect. Tightly integrated, high density and power efficient it probably isn't, but then again neither are most of the other scale out solutions it's competing with, it's just they tend to scale much further..

It's just a theory but EMC make a lot of noise about everything being handled in RAM (always on etc), hence the need for the external UPS devices, but I wouldn't mind betting the RAM size is what actually limits the addressable capacity and the more services you add the more metadata will need to reside in RAM.

I hear they still can't scale out on the fly but they will provide swing kit in the form of another system so that you can Svmotion to another unit while they upgrade the old and then Svmotion back (assuming you are exclusively VMware). Maybe that's where the growth numbers come from, all the swing kit they had to buy internally :-)

[Man Mountain]

The bigger these claims of capacity savings become, the more the scope for a significant mis-sizing! As we all know, 'your mileage will vary' so the bigger that claim, the bigger the margin for error! Given the skepticism that even still exists in some corners around technologies like thin provisioning, I can imagine many IT shops having significant concerns over sizing, especially when close to the upper limit for these relatively small AFAs.

[Texas]

POC

I'm pretty sure EMC offer POC's, easy way to test out these things.

[Anonymous Coward]

Re: POC

As will pretty much any other Vendor