Brad Templeton [EMAIL PROTECTED] wrote:
> 
> Top can be misleading with HT.  But my machine is also HT and it's only
> 60% idle, not 80%.  And it's faster.   Do you have dual channel RAM?
> Can that be making this big a difference?  Can your video card be that
> much faster?  Could Mandrake be doing much better than FC3?

If you run the binary Fedora kernels, then one possible factor could
be the 4G+4G address space patch that Fedora kernels are built with.
This turns each copyin/copyout from a simple bcopy with a trap handler
into bcopy + a complex vm remapping.  This increases the cost of
system calls, especially "simple" calls like gettimeofday, small
reads, etc, because the cost of the copyin/copyout now dominates
anything else that may be happening.  This would not be much of a
factor for large reads, for example.

On a 3.06GHz P4 HTT system I have here, running stock a FC2 kernel
(vmlinuz-2.6.5-1.358smp) lmbench's lat_syscall shows 2.4us for a "null"
(getpid) syscall, and 3.25us for a read syscall.  Compare this to a
a 2.6.5 SMP kernel.org kernel with the traditional 3G+1G address space
on the same machine, which shows 0.38us for a "null" syscall, and
0.46us for a read syscall.

Admittedly this may not amount to much in the scheme of things, or
things may have gotten better with FC3.  I also do not know how often
myth makes syscalls, as I'm waiting eagerly for the UPS man to deliver
the components of my first box.

Drew



_______________________________________________
mythtv-users mailing list
[email protected]
http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users

Reply via email to