So a friend of mine (@vaughanjt) tweeted that out to me today:
@aus_effendi I know you're anti-host storage, but damn….what throughput. pic.twitter.com/oTivuuGliz
— JT (@vaughanjt) September 5, 2015
and I responded
@vaughanjt not against host storage at all it's just my current issues are VM management (cloning, replication, etc) vs IOPs performance
— aus_effendi (@aus_effendi) September 5, 2015
@vaughanjt so solving those issues is top goal – getting VMs the IOPS they need also important, but my IOPS needs pretty modest
— aus_effendi (@aus_effendi) September 5, 2015
@vaughanjt and finally, using RAM (like @PernixData ) will be even faster
— aus_effendi (@aus_effendi) September 5, 2015
At my employer, we’ve got around 500 VMs in one location, around 300 of which are various builds of the software we produce (vs back office stuff like Exchange, Oracle, etc). These 300 VMs are usually 10-30 copies of the same master with some small networking changes. At present time on our existing storage (EMC VNX 5500), doing those 10+ copies of the master are my bottleneck.
Yes, going to some sort of AFA (All Flash Array) could make this operation take alot less time. Yes, implementing something like vCAC / vRA (vCloud Automation Center / vRealize Automation) could definitely help with that bottleneck – linked clones off the master and all that, but simply, with my recent promotion and my need to backfill three personnel in the Infrastructure and Operations group, I simply don’t have the time to get that implemented in a robust and redundant way right now.
Couple that with other business challenges we face (the lease / maintenance on the VNX is due end of year, replication, etc among many other factors), and the best solution for my employer, based on our current constraints, was new shared storage that can provide the IOPS I need, yet greatly simply / speed up the management operations that are a bottleneck in our build cycle.
Host based storage (and IOPS in general) is a great thing, if it solves the problem you’re really facing. If, after solving the existing bottlenecks, I need more IOPs, I’m all for using host based storage (RAM, NVMe, SSD, etc) for caching to get me the IOPs.
Flash can be the proverbial hammer to solve a problem, but sometimes re-architecting your process is a better solution than throwing more IOPS at a problem.