Hi all, I'm running the data caching benchmark and measuring TLB miss rate for the memcached server (via hardware performance counters).
I get very low TLB miss rates, around 0.1 - 0.2 %, which seems unrealistic to me. How to scale up the workload and get higher TLB miss rates? So far I tried: 1) scaling up the input twitter dataset from 10GB to 20GB had no effect at all (probably because it preserves the keys popularity). 2) specifying larger object size (e.g. fixed object size of 64 KB) had a very minor effect (TLB miss rate moved from 0.15% to 0.2%). Maybe I don't fully understand the memcached behavior, so I'll appreciate any help. Thanks!
