Dear Cache Guys,
Is it possible for my cache to be too big?
As we learned previously, one popular method of sizing the cache is to make a best guess at the I/O working set with knowledge of the workloads running on an ESXi server, observe results for a period of time to see how it is working, then adjust as necessary.
One concept to consider where on the cache hit ratio curve the combination of your working set and cache size falls. Consider the graph below, which shows typical hit rate behavior when increasing cache size. Initially, increasing the cache size creates large gains in cache hit rates. But eventually adding more does little or nothing, which ultimately does not improve system behavior.
Determining where the inflection point is on this curve can be a challenge when tuning your cache, but a critical one. While we all can agree that a cache that is too large is better than one that is too small, there can be drawbacks to a cache that is too large as well. For example, in a RAM-based cache like Infinio, you may have reached a point of diminishing returns, and have memory dedicated to the cache that could be used by other processes.
The good news is that Infinio provides excellent tools for evaluating the results. The Infinio dashboard reports cache effectiveness allowing an administrator to quickly understand how the cache is performing. And Infinio makes it simple to resize the cache when adjustments should be made, not requiring reboots or downtime. It is still half-art and half-science, but the Infinio tools provide clarity into what is happening.