When running a YCSB 50/50 workload with 16 threads throughput can decrease significantly for around 10 seconds corresponding to a portion of a checkpoint in WiredTiger.
Investigate the root cause of the reduction in throughput, and determine if there is something we can do to mitigate it.
The root cause turned out to be that we disabled lookaside eviction while a checkpoint was running - which meant that updates caused the cache to get stuck.
The fix was to allow lookaside eviction while checkpoints are running and to restructure checkpoint to minimize the amount of content read from the lookaside file.