[SERVER-16355] Extended pauses with heavy write load under WiredTiger Created: 30/Nov/14 Updated: 18/Dec/14 Resolved: 15/Dec/14 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | Storage |
| Affects Version/s: | 2.8.0-rc1 |
| Fix Version/s: | 2.8.0-rc3 |
| Type: | Bug | Priority: | Major - P3 |
| Reporter: | Bruce Lucas (Inactive) | Assignee: | Alexander Gorrod |
| Resolution: | Done | Votes: | 0 |
| Labels: | wiredtiger | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Attachments: |
|
| Backwards Compatibility: | Fully Compatible |
| Operating System: | ALL |
| Participants: |
| Description |
|
Under heavy write load from many threads, there are occasional several-second pauses where the insert rate drops to 0. Test consists of standalone mongod with 100 threads of the following:
Top four graphs below are derived from wt statistics and mongod log. Remainder is call tree obtained by sampling with gdb; graphs next to each call site show number of threads at that call site over time (on log scale).
Reducing memory_page_max from the default of 100MB to 10MB eliminates the extended pauses. Turning off compression also reduces the length of the pauses. The call stack seems to indicate that in this particular test (on a machine with SSD drives) this is all CPU time, as no i/o operations show up in the gdb stack samples. So it appears that
|
| Comments |
| Comment by Bruce Lucas (Inactive) [ 12/Dec/14 ] |
|
Yes, it is considerably improved. rc1: rc2:
|