CBRC has been a great addition to the VMware architecture and offer's some real benefits to those looking to offload storage requests without additional products or costs. Since the most common blocks are cached in memory and there is no need to request those blocks from the storage infrastructure, there is a reduction in overall storage traffic particularly during boot storms.
The following test data showing the reduction in IOPS and storage bandwidth was posted by VMware on their End User Computing Blog:
However, View Storage Accelerator does have some limitations:
1.) It's a read only caching technology.
The write IOPS still have the same heavy impact as before. If you take away boot storms, write IOs are usually estimated to be between 50-80% of the total IOs during steady state virtual desktop operation.
2.) It's limited to 2GB of RAM per host.
It is a dynamic cache, but in today's age of relatively cheap server DRAM, this is starting to seem a little limited. I'm sure 2GB was determined at one time to be a maximum "sweet spot" for resource usage versus benefits gains. However, since I run primarily linked clone non-persistent desktops in my environment, I wonder how fast a desktop could be if say 16GB of the desktop's main replica image could be cached in RAM.