Re: Lock profiling results on TCP and an 8.x project

On 10/21/07, Robert Watson <rwatson@xxxxxxxxxxx> wrote:
On Sat, 20 Oct 2007, Robert Watson wrote:

* When talking about percentage of available CPUs, I make the assumption
that
due to a sufficient quantity of CPUs, in most cases lock acquisition will
occur as a result of adaptive spinning rather than sleeping. In the
netperf
case, this is not true, since the number of potential workers exceeds the
number of CPUs, hence the turnstile contention. However, as sleeping on
locks itself is very expensive, it's reasonable to assume we would recover
a lot of CPU none-the-less.

FYI, a feature request for lock profiling: it would be nice if we also tracked
for each contention point time spent spinning vs. context switched waiting for
the lock, and the number of context switches the lock acquisition point has
caused. This would allow us to better understand the impact of adaptive lock
behavior for workloads and configurations.

This would involve changing acquisition wait time into 3 fields:
- time spent spinning
- time spent context switched out
- number of context switches

The lock profiling object is fairly large as it is, so ideally this
would happen after I've moved the lock_profile_object out of the
lock_object. That is work that I've already mostly done as part of
making profiling work for shared acquisitions and having it not break
the ABI when enabled. Unfortunately, I haven't had time to make it
commit-ready - so if someone wants to jump in and implement Robert's
idea before that happens I won't object.

-Kip
_______________________________________________
freebsd-arch@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-arch
To unsubscribe, send any mail to "freebsd-arch-unsubscribe@xxxxxxxxxxx"