[MCOL-3460] experiment with larger recv buffers in writetask/appendtask Created: 2019-08-26 Updated: 2021-01-26 Resolved: 2019-09-03 |
|
| Status: | Closed |
| Project: | MariaDB ColumnStore |
| Component/s: | ? |
| Affects Version/s: | None |
| Fix Version/s: | 1.4.0 |
| Type: | Task | Priority: | Major |
| Reporter: | Patrick LeBlanc (Inactive) | Assignee: | Ben Thompson (Inactive) |
| Resolution: | Done | Votes: | 0 |
| Labels: | None | ||
| Description |
|
Noticed that writetask and appendtask are breaking incoming data into 1MB chunks to write. That made sense for milestone 1 when those were written, but might make sense now to try to do the whole write op in one IOC call rather than break it up. Need to experiment and see if it's worth the extra mem usage. |
| Comments |
| Comment by Patrick LeBlanc (Inactive) [ 2019-08-27 ] |
|
Found that this was worth doing. Locally using test000 as the benchmark, got an end-to-end gain of ~7%. Raised the cap to 100MB for safety. Should be 1/100th the impediment it was, and keep ridiculous things from happening. |