[SERVER-16662] Extended pauses in WiredTiger when workers greatly exceeds number of cores Created: 24/Dec/14 Updated: 24/Apr/15 Resolved: 23/Mar/15 |
|
| Status: | Closed |
| Project: | Core Server |
| Component/s: | Storage, WiredTiger |
| Affects Version/s: | None |
| Fix Version/s: | 3.0.0 |
| Type: | Bug | Priority: | Major - P3 |
| Reporter: | Bruce Lucas (Inactive) | Assignee: | David Hows |
| Resolution: | Done | Votes: | 0 |
| Labels: | wiredtiger | ||
| Remaining Estimate: | Not Specified | ||
| Time Spent: | Not Specified | ||
| Original Estimate: | Not Specified | ||
| Attachments: |
|
||||||||||||||||||||
| Issue Links: |
|
||||||||||||||||||||
| Backwards Compatibility: | Fully Compatible | ||||||||||||||||||||
| Operating System: | ALL | ||||||||||||||||||||
| Participants: | |||||||||||||||||||||
| Description |
|
This is a continuation of
Under heavy mixed workload with 100 clients, periodic extended pauses of several seconds with zero throughput are seen, for example at A and B:
The length of the pauses is roughly proportional to the number of threads, supporting the theory that the CPU contention due to the spin lock exacerbates the pauses. Here's the same 200 second run with 50 clients, showing pauses roughly one half the length of pauses with 100 clients above.
Similar symptoms are seen in a less extreme form with a 100-thread insert-only workload:
Insert-only workload:
|
| Comments |
| Comment by Michael Cahill (Inactive) [ 13/Jan/15 ] |
|
This workload should be considerably improved by RC5. |
| Comment by John Page [ 09/Jan/15 ] |
|
Just seen/realised this is especially pertinent in a sharded setup as each monogs router will make a whole bunch of connections and thus threads - I'm seeing as I add routers - overall performance drops in 2.8 - However! it also does with mmapv1 and it doesn't in 2.6 so this is more than simply wired tiger related. |