This article is more than 1 year old
EMC morphs storage arrays into servers
Quacks like a duck, runs apps like a duck
Comment EMC is extending its storage arrays into application engines.
Oracle and others, such as start-up Nutanix, have had the idea of bringing stored data closer to the server to reduce network latency and get faster access to data.
An additional way of getting much faster data access is to put a load of flash memory into servers, which is what Fusion-io is doing with its ioDrives. Another approach is the HP OEMing of Violin Memory flash arrays, where it is hooking them directly to servers with PCIe connections, providing up to 80TB for a single server.
Let's put this flash/server idea off to one side for a moment of pedantry.
Users don't really want the data brought closer to the server. The server is after all merely a resource for running applications - what is needed is to bring the data closer to apps. Bear this in mind as we turn to EMC.
Why would EMC bother? It's not in the server business ... Oh but it is.
EMC wants to enable the close coupling of data storage and servers (applications): but the storage giant approaches this from the other direction: bringing the apps to the array and running Intel array controller engines, otherwise known as servers. This has been mentioned before by El Reg as a logical speculation, partly based on hints from EMC staffers.
Now an EMC staffer, Mark Twomey, has blogged the curtains wide open. This is what he said about demos at VMworld:
If you wanted to be "sensitive" ... in how you stated it you'd say 'EMC is now in the Compute business'. Or it will be when we ship the feature.
EMC has been demoing moving Virtual Machines from other people's servers to running directly on the storage array: twice on stage, a lot more than that behind closed doors. Isilon are showing demos of it running in public, the VMAX team have started talking about it in public and there are other shoes to drop.
Just assume that in the future, when it's easier to move the Compute instead of the Data, you'll be able to move the Compute directly onto the EMC Array.
Is that being in the Server business? Maybe. But it's more like being in the Systems business.
Let's return to the flash/server idea. In a Silicon Angle article, EMC president Pat Gelsinger is quoted as saying that EMC is catching up with Fusion-io. Really? The Fusion-io that provides server flash? Why would EMC bother? It's not in the server business ...
If it walks like a duck and quacks like a duck ...
Oh but it is. EMC will run applications inside VMware virtual machines, in spare VMAX engines and Isilon controllers – and as Twomey says, these are servers in all but name. EMC is in the server business and it is going to accelerate these servers with flash cards and it will, El Reg believes, run those flash cards from the array as part of the Project Lightning technology thread.
Here is the Gelsinger quote from Silicon Angle: "We took [an array] race-centric view and we're a leader in applying Flash in the array. What we didn't do is say our business is about Flash in the array – we're about storage no matter where it's at. We're out to create different value of Flash in the infrastructure, and that will give us a differentiated position in the market."
Okay, so EMC will make integrated storage array/flash-enhanced server combinations that will compete with server-plus-storage array combinations from Dell, HP, IBM and Oracle. Will Dell, IBM and HP arrays be able to run applications?
If not they will be at a disadvantage.
Let's push this thinking one step forward. If you have a VMAX, with flash-enhanced engines, able to run application software, then you wouldn't need UCS servers to do that job. Were EMC to do a deal with a network supplier, then you wouldn't need Cisco network switches to hook the application server/array complex up to accessing clients either, and we might have a VMAXblock as well as a Vblock. ®