In DSC, we are aiming to complete a checkpoint once every five seconds. That value was chosen to help manage failover times and standby staged work volume.

There is some leeway in how frequently checkpoints need to complete. Five seconds was chosen as a conservative starting point, but we should experimentally see how much cache space is required to keep a copy of changes between checkpoints. If there is enough space to manage more than five seconds worth of load for the YCSB 100 update and 128 thread load cases, we could reduce checkpoint frequency to 7 or 10 seconds.

Fewer checkpoints will cause less data to be written back to SLS, which will allow for higher throughput in these update heavy workloads.