After a memcached restart Magma's performance is 15-20% higher in a read only YCSB C workload at 1% dgm. This degrades over time and stabilises in about an hour. This is mostly caused by je_malloc fragmentation building up.
We have to investigate and confirm the cause. Also see if it can be mitigated for low dgm zipfian usecases where we have a 50% cache hit ratio and small changes in memory usage can cause large deviations in throughput.