At the moment our vSphere system is running against 3 HP EVAs. Two of these are in year 4 of their life, one is new. Question is should we look to replace our older 8100’s with a new 6400/8400 unit. Is this the most cost effective solution, and is it even a good idea to run vSphere against a single monolithic storage system?
Cost wise, it turns out that buying a new EVA with similar capacity to the older 8100’s is about the same as 3 years maintenance for the older units. The downside is that we spend money on disks during the last three years, and these won’t fit the new units, so we scrap (or resell) the old units with disks that are less than a year old – not really good value or very green!
From an eggs in baskets perspective, we have all our production VMs – heading into the 100s of VMs dependent on just two controllers in a single EVA. As we have experienced in the past a single controller failure (or even a disk unit) can bring down the whole unit. So we’ve migrated from lots of physical machines with their own independent storage to lots of VMs sharing a single unit with (despite HP’s protestations) a single point of failure!
So should we be moving away from this architecture towards having a larger number of independent baskets, each containing few eggs. We’ve started on this path for disk based backup using HP P4000 (LeftHand) units which do exactly this, but present the storage of a number of physical units as a single storage array. Clearly, it may still be possible for a single member of this array to fail in such a way as to kill the whole array, but also it’s clearly possible to run several arrays in parallel, so balancing the risks against the pain of management – a small number of baskets, each containing a number of baskets which look after a subset of your Eggs – OK so analogy failure here, but hope you get the picture. One key advantage of this is upgrade costs, no longer do you “forklift” in a replacement EVA, you can add units over time, and retire older units over time as budget allows.
After a while of course you get too many baskets to make life easy to manage – which VMs on which storage array, but is this much worse than which VM’s on which LUN? There are obviously solutions to this by virtualizing the storage, but this usually involves introducing an expensive physical unit to perform this work, well two would be needed to fault tolerance, but then you’re back to the EVA with just two controllers managing all your storage!
So an approach along the lines of the HP P4000’s should be the right one for us, but do we look at other options, XioTech, Equalogic, etc? Do we throw in clever units like Nexenta produce to make better use of the storage we have, but again introducing some additional points of failure?