CloudPhysics continually surprises me with their innovation when providing cards for simulation purposes. I’ve posted a couple of times already about how I really like their AWS pricing calculator (they also have vCHS as well).  Having a good idea about how much your existing environment will cost if you make modifications is a pretty big win for a CIO.

While I was at GestaltIT’s Virtualization Field Day 3, two weeks ago, Irfan Ahmed showed us a new card that would simulate how much SSD Cache you should buy based on your current workloads.

All travel expenses and incidentals were paid for by Gestalt IT to attend Virtual Field Day 3. This was the only compensation given.

If you’ve been living under a rock for the last 18 months, solid state drives are transforming how we look at storage.  An entire industry has been spawned with the sole goal of speeding up traditional spinning disks by putting some form of SSD in front of them in the IO path.  PernixData has a product called FVP and VMware has released vFlash just to name a couple of products that use host based caching with SSDs.  Tintri, Nimble Storage, and Coho Data are among some of the newer storage arrays that were built with SSD Caching in mind right from the start.  The big storage arrays like Netapp and EMC have also added SSD caching to their arrays as well.

Clearly writing data to the SSD is faster than waiting for a mechanical disk, but the question then becomes, “How much cache do we need?”  The idea of cache is to use small amounts of SSD for performance but to keep the spinning disks in your environment for large bulk storage.

Irfan explained that CloudPhysics uses your own workloads to determine where that sweet spot is.  He used a graph similar to the crudely built graph below that I mocked up for an example.

If we look at IOPS on the left and GB’s of cache along the bottom, we’ll notice that in most cases just adding more cache doesn’t help.  This was a wake up moment for me.  Cache is good, but more cache is better right?  Well, not necessarily.

This crudely drawn graph is just an example and in no way should reflect upon CloudPhysics.  🙂

You can see in the graph that as the size of the cache increases, the IOPS increase, but there are jumps where the amount of cache really helps.  For instance the difference between having 30 GB of cache vs 40 GB of cache only improves the IOPS slightly.  However, going from 45 GB to 50GB dramatically increases performance.  This is due to the fact that the working set may be thrashing and evicting data from your SSD.  If you can fit the whole working set into the SSD performance should dramatically increase.

Summary

This seems like a pretty simple concept, but until Irfan mentioned it, I’d never thought about it before.  If you’re considering buying cache for your environment, CloudPhysics might have a tool that can help you determine how much cache to get without paying for useless space.