jhouserizer
commented
1 year ago Ehcache uses a tiering system. Yes, an entry that was just read from the cache is likely to be promoted in to an in-memory tier (offheap and/or heap).
Additionally, your OS is likely caching your file system, so a second lookup is likely to be faster even if from disk.
Lastly, 2 seconds is an eternity. Guessing that the disk is maybe virtual/network based? Are you also measuring JVM warm-up (JIT) time?
I was trying ehcache as a disk tier cache only option and running some benchmarks. I ran multiple iterations to fetch a specific key from ehcache and observed high latency from first cache hits when compared to the rest of the iterations. Something like(doing multiple cache gets with same key): 2sec(cache_miss_latency), 30ms(first cache hit), 1ms, 2ms, 1ms ....... so on.
I cached bunch of data and was trying to get a specific key multiple times.
I wanted to know why this behaviour ie first cache hit is expensive, is it because it reads from disk for first time and promotes to offheap for later retrievals?
Also I deleted the folder where ehcache data is stored and it was still able to serve reads and writes(surprisingly). Need an explanation for this as well.