Subject: Re: Thread benchmarks
To: None <firstname.lastname@example.org>
From: Andrew Doran <ad@NetBSD.org>
Date: 10/03/2007 14:23:55
On Tue, Oct 02, 2007 at 10:21:53AM -0700, email@example.com wrote:
> In message <Pine.NEB.4.64.0710012230280.900@S.culver.net>,
> Mike Cheponis writes:
> >On Fri, 28 Sep 2007, Andrew Doran wrote:
> >> On Fri, Sep 28, 2007 at 06:50:16PM +0100, Andrew Doran wrote:
> >>> On Fri, Sep 28, 2007 at 02:27:34PM -0400, Thor Lancelot Simon wrote:
> >>> Either we degrade a lot more gracefully than Linux under load, or there's
> >>> an artifact in the Linux graph. The current plot makes it impossible to
> >>> tell which, though.
> >> (...stuff removed...)
> >> In the long run Linux will beat NetBSD.
> >Why is this?
> >I don't know any fundamental reason this should be so.
> There are at least two ways to take Andrew's quoted text. One way is
> that over time, Linux will do better than NetBSD (in the long haul,
> which is how I first read the quoted line). The other way is that at
> this point in time, if we look at datapoints beyond the right edge of
> Andrew's graphs from last week, Linux does better than NetBSD. From
> off-list discussion with Andrew, I am sure he means the second, not
> the first.
> Andrew's comments in another message, referring to a gradual drop-off
> with increasign number of connections and suggesting kernel profiling
> in that regime, to find the source of the gradual drop-off, also
> support thet second reading.
> Andrew can say more on this score if he chooses.
Right, it was badly worded. With the later peaks that Linux shows, and with
NetBSD's gradual fall off, continuing on to (say) 500 threads will show
Linux achieving a higher transaction rate.