EMC's flashy XtremIO? I Xpect it will be great... in a few years

So we finally have general availability for XtremIO – not that general is much different from directed availability: it is still going to be pretty hard to get hold of an XtremIO if you want one. And that is of course a big *if*: do you really need an all-flash array? Can you use it? Is it just going to be a sledge-hammer to …

[klaxhu]

Too early to say...

The author makes is sounds like only Xtrem IO from all the AFA's (Pure, Violin, etc) has this problem of being "enterprise ready" where the reality is they all have the same problems and it's normal when you design something from ground up. The reason why nobody is rushing to have those data services is because they will all hit performance in a way, so they will probably let somebody else do that at a different level. In EMC's case I am guessing that technologies like VPLEX will be used to provide those services...

Is the market ready for AFA's? Probably not, but it is definitely good to have a look and separate your workload into high performance and high availability because of how these products change the market-place. Today the VMAX covers both, but if you look at most customers who buy a VMAX, they buy it for the availability and not the performance as it was wrongly pointed out many times on The Reg. in the comment fields.

PS: yes, I have created this account just to post this, but I have been following the website and the discussions here for quite a while now ...

[smt789]

Depending upon the AFA, there are other value props for leveraging the technology than just performance. Some incorporate the promise of what DataDomain wanted to do prior to EMC buying them - inline dedupe and compression, etc. This makes them very space efficient which in turn impacts power, data center footprint, etc. You can buy X amount raw but get X*3-5 usuable. There is management as well. There are some neat implementations of how you manage these systems. Others not so much though.

I see the workloads for these being OLTP. Those apps drive the business. And from a vendor perspective, that is where the money is. VDI and VMware a good plays, but OLTP is where they should play. To the author's point, how enterprise ready are the AFAs? Replication is key but there are alternatives - namely app level replication for OLTP (DataGuard, etc.). That said to be truly enterprise ready, they need true sync and async replication with consistency groups.

[Anonymous Coward]

The main use for enterprise flash is workloads with many small read cache misses. These are the I/Os which are forced to fetch data from underlying storage, as opposed to cache. Applications with I/O heavy workloads will benefit, particularly those whose I/O can't be parallellised. If your workload must be sequential (this is random reads happening sequentially, not sequential I/O) and your read cache miss response time is 20 ms (not unusual for spinning disk) then you're going to get 50 IOPs, which is useless. Reduce that response time to 1 ms, and that number goes up to 1000. Reduce it down to 100 us, which the best flash storage seems to be able to do currently then you're more likely to move the saturation point to the server before you reach 10000 IOPs.

If your workload can do many things at the same time, and you have a decent queue depth, or you have few cache misses, then flash is probably a waste of money.

As for this EMC offering, it seems to be able to do not very much, not very well. It seems like this GA was purely dictated by marketing. The product has nothing new or better to offer than other more established offerings. No doubt EMC's salesmen will bang on about how many years they've had the lion's share of the storage market, but it's about as relevant here as if they were selling me a wardrobe.

[Michael Duke]

XtremeIO is a tool in EMC's kit bag, strategically not as an individual offering (Not that they will turn down sales I am sure, but even EMC are pitching it at big VDI right now) but as part of a complete "Big Picture" play with ViPR providing the "Software Defined Storage" control layer with XtremeIO, VMAX, VNX, Isilon, ScaleIO and Data Domain as storage resources to be managed and VPLEX/RecoverPoint providing the availability piece.

It is an impressive story that really only IBM can compete with, but even IBM have some work to do in this space.