The paper that I mentioned last week, that contains some FreeBSD
performance data, is about Hourglass, a user-level tool for measuring
scheduling behavior. It'll appear in the FREENIX track of the USENIX
general technical conference in June.
There's a draft here:
http://www.cs.utah.edu/~regehr
> > > Have you tried benchmarking process to process context switch times to see
> > > if the results are similar?
I put a quartet of histograms here:
http://www.cs.utah.edu/~regehr/freebsd_ctx_quantum.eps
http://www.cs.utah.edu/~regehr/freebsd_ctx_yield.eps
that demonstrate, basically, wha
> The problem is that it's not clear what the graphs you posted
> are comparing. In the context of the paper, this will probably
> be mitigated somewhat. However, there are a lot of people who
> will turn directly to the graphs in any paper, and yell about
> them, so I doubt you are safe, not ma
Stephen J Bevan wrote:
> This is the same way that the scheduling CPU and process group
> affinity crap that Linux puts up with just falls out of the
> code, as well, when you go to per CPU run queues
>
> Since the thread on freebsd-arch didn't appear to have anything
> specifically to do w
John Regehr wrote:
> No need to use me as an excuse to vent your feelings about
> microbenchmarks vs. good benchmarks. I'm showing how to use a
> user-space instrumented application to measure scheduling behavior, not
> trying to make any claims about the relative merits of the operating
> system
Stephen J Bevan wrote:
> > The correct approach for CPU affinity is to run with per
> > CPU scheduler queues. ...
>
> If "scheduler queue" means the same as "run queue" then
> per CPU run queues were added to Linux in 2.5.2. See
> http://lwn.net/2002/0110/a/scheduler.php3.
I know this. The v
On Thu, 4 Apr 2002, Zwane Mwaikambo wrote:
> On Wed, 3 Apr 2002, John Regehr wrote:
> > There does not appear to be a statistically significant difference
> > between a native binary and an emulated Linux binary.
>
> Wouldn't the only place you'd notice slight overhead be syscalls?
>
> Zwa
On Wed, 3 Apr 2002, John Regehr wrote:
> > > Have you tried benchmarking process to process context switch times to see
> > > if the results are similar?
>
> No, that's a good idea. My infrastructure isn't set up to support
> processes, though, so it'll take a little time.
I'd be surprised if
Terry,
No need to use me as an excuse to vent your feelings about
microbenchmarks vs. good benchmarks. I'm showing how to use a
user-space instrumented application to measure scheduling behavior, not
trying to make any claims about the relative merits of the operating
systems in realistic condit
John Regehr wrote:
> I'm writing a paper for the FREENIX track at USENIX about a tool for
> measuring scheduling behavior under different operating systems, and
> I've run across a performance anomaly that I'd like to explain.
>
> It's this: for threads created with Linuxthreads, FreeBSD has
> co
On Thu, 4 Apr 2002, Zwane Mwaikambo wrote:
> On Wed, 3 Apr 2002, John Regehr wrote:
>
> > > > Have you tried benchmarking process to process context switch times to see
> > > > if the results are similar?
> >
> > No, that's a good idea. My infrastructure isn't set up to support
> > processes, t
> > Have you tried benchmarking process to process context switch times to see
> > if the results are similar?
No, that's a good idea. My infrastructure isn't set up to support
processes, though, so it'll take a little time.
> Also:
> You should run both linuxthreads binaries compile on linux
On Wed, 3 Apr 2002, Mike Silbersack wrote:
>
> On Wed, 3 Apr 2002, John Regehr wrote:
>
> > Anyway, I was speculating that the higher cost is either due to (1) a
> > failure, in FreeBSD, to avoid page table operations when switching
> > between threads in the same addres space, or (2) some ot
On Wed, 3 Apr 2002, John Regehr wrote:
> Anyway, I was speculating that the higher cost is either due to (1) a
> failure, in FreeBSD, to avoid page table operations when switching
> between threads in the same addres space, or (2) some other kind of
> semantic mismatch between Linuxthreads and r
Hi all,
I'm writing a paper for the FREENIX track at USENIX about a tool for
measuring scheduling behavior under different operating systems, and
I've run across a performance anomaly that I'd like to explain.
It's this: for threads created with Linuxthreads, FreeBSD has
considerably slower cont
15 matches
Mail list logo
