Messages in this thread | | | Subject | Re: [PATCH -v4 0/8] locking/mutex: Rewrite basic mutex | From | Jason Low <> | Date | Tue, 11 Oct 2016 11:42:41 -0700 |
| |
On Fri, 2016-10-07 at 16:52 +0200, Peter Zijlstra wrote: > Hi all, > > Since you all should not be sending patches during the merge window, I figured > I should to keep you all occupied with something. > > Please review, test and otherwise try to break these here patches. > > I would like to get these patches into -tip (and -next) once the merge window > closes, so please spend these quiet days staring at this stuff. > > Small changes only, mostly the handoff logic as suggested by Waiman last time.
I tested these new patches on the 8 socket system with the high_systime workload which stresses mutexes.
The average throughput with and without the patches were very similar at lower levels of contention. At high contention, the throughput with the patches were slightly lower, but only by a small amount, which I think is expected due to the additional waiter wait time, ect...
------------------------------------------------- | users | avg throughput | avg throughput | | without patch | with patch | ------------------------------------------------- | 10 - 90 | 13,989 JPM | 13,920 JPM | ------------------------------------------------- | 100 - 900 | 76,362 JPM | 76,298 JPM | ------------------------------------------------- | 1000 - 1900 | 77,146 JPM | 76,061 JPM | -------------------------------------------------
Tested-by: Jason Low <jason.low2@hpe.com>
| |