• 64 GB memory, 24 cpus (12 cores)
• mongod 3.3.5, standalone, 45 GB cache
• YCSB workload as follows (in file repro.ycsb):

  fieldcount=3
  recordcount=800000000
  operationcount=400000000
  workload=com.yahoo.ycsb.workloads.CoreWorkload
  
  readallfields=true
  
  readproportion=0.5
  updateproportion=0.5
  scanproportion=0
  insertproportion=0
  
  requestdistribution=zipfian
  

  run as follows:

  ycsb load mongodb -s -P repro.ycsb -threads 30 -p batchsize=1000
  

Some key metrics from a run:

• stall from A-B immediately followed by
  • large jump in pages held in cache
  • very large maximum page size at eviction (about 10 GB)
  • immediate large jump in excess allocated memory of about the same size (11 GB)
  • excess immediately begins to decline
• other stalls, e.g. C-D, appear to be associated with cache exceeding configured maximum (49 GB in cache vs 46 GB configured)