[SERVER-78319] Investigate execution control tuning effects Created: 22/Jun/23 Updated: 06/Sep/23 Resolved: 31/Jul/23 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | None |
| Affects Version/s: | None |
| Fix Version/s: | 7.1.0-rc0 |
| Type: | Task | Priority: | Major - P3 |
| Reporter: | Gregory Noma | Assignee: | Shin Yee Tan |
| Resolution: | Done | Votes: | 0 |
| Labels: | RDY, server-rapid-response-resolved | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Issue Links: |
|
||||||||||||
| Backwards Compatibility: | Fully Compatible | ||||||||||||
| Sprint: | Execution NAMR Team 2023-07-10, Execution NAMR Team 2023-07-24, Execution NAMR Team 2023-08-07 | ||||||||||||
| Participants: | |||||||||||||
| Description |
|
While for 7.0 we plan to keep the current default values for execution control tuning parameters, we should understand and document the effects of tuning these parameters. In particular, we've seen that when the system is beyond capacity for both reads and writes, the mix of reads and writes may be different with execution control enabled. The throughputProbingReadWriteRatio parameter will likely be important here. |
| Comments |
| Comment by Shin Yee Tan [ 31/Jul/23 ] |
|
In linkbench and ycsb workloads, we observe no significant change from enabling execution control and from toggling different throughputProbingReadWriteRatio for regular system loads. For mixed_workloads_genny_stress where we run 1024 threads with point lookups, finds, inserts, deletes, and updates: we observe a significant improvement in write performance at the expense of a small regression in read performance. With mixed 10k thread workloads, we test our system under load. When compared to performance without execution control enabled, we observed improved write performance (increased throughput and decreased latency) across all 3 workloads no matter what the readWriteRatio was. We see the greatest increase in write throughput with readWriteRatio at 0.5 and smaller increases as readWriteRatio was tuned higher. When the 10k threads are split evenly between reads and writes, we see regressions in read performance with readWriteRatios ranging from 0.5-0.95 however we see smaller regressions with higher readWriteRatios. For 7.5k reads 2.5k writes, we saw an improvement in read performance when readWriteRatio was increased to 0.9. For 9k reads 1k writes, we saw an improvement in read performance when readWriteRatio was tuned up to 0.6 and beyond. |
| Comment by PM Bot [ 22/Jun/23 ] |
|
This issue has been flagged for rapid response! Assignees of rapid response tickets are responsible for providing a daily update on this issue using the 'Server Rapid Response' canned comment template. Any questions about this ticket can be directed to the #server-rapid-response Slack channel and more information on the Server Rapid Response process can be found on the Wiki |