Subject: Re: i386 MP performance (generic.mp and generic almost same)
To: Cem Kayali <cemkayali@eticaret.com.tr>
From: Steven M. Bellovin <smb@cs.columbia.edu>
List: current-users
Date: 11/11/2007 03:47:17
On Sat, 10 Nov 2007 18:06:11 -0800 (PST)
Cem Kayali <cemkayali@eticaret.com.tr> wrote:
>
> Hi,
>
> I would like to ask experiences of netbsd-current-users running
> multi-procesor kernel. I have core-duo notebook, and am using standart
> generic.mp kernel. dmesg shows i have cpu0 and cpu1.
>
> Tonight, i have compiled generic kernel, and noticed that there is
> almost no performance difference... I mean,
>
> - Boot time is almost same (untill kde desktop just loaded)
> - KDE performance is almost same (windows startup etc)
> - Linux emulation software (Firefox,OO2) peformance (startup time) is
> same.
> - More importantly, compiling a new kernel or compiling a software
> through pkgsrc (from source) time is almost same.
>
>
> Well, i think i should somehow notice difference between generic and
> generic.mp kernels...
>
Multiprocessor performance will only be different when there are two or
more CPU-bound processes competing for the CPU. Boot time is, I
believe, dominated by I/O (and in particular disk) performance; the
same is true for starting up large applications such as Firefox.
Compilation is indeed CPU-bound; however, unless you've taken special
steps, it's a single process at a time. What you need to do is specify
'-j 2' when invoking build.sh, to run two processes simultaneously;
see /usr/src/BUILDING for details.
pkgsrc can only benefit from that if the Makefiles in the packages are
written appropriately. I suspect that virtually none are.
--Steve Bellovin, http://www.cs.columbia.edu/~smb