W Bauske wrote:
> > As a last resort, if one tires of seeing those messages, one could up
> > the default "timeout" value in the debugging locks code to wait for a
> > longer time before giving the message... :-\
> >
>
> I don't look at the system console much unless there's trouble
> or I need to shut things down so I rarely look at these messages.
> The key is they are not specific to boot time. I take it these
> messages indicate the hang is around 2 seconds, which makes sense
> given I send 16MB over 100Mb Enet, which would mean PVM is sustaining
> around 8MB/sec, actually quite good from my point of view. I'm
> surprised the other processor would have to wait while another
> processor was receiving network data. The lock must protect large
> portions of the kernel code. At some point I'll go back and try
> the 2.4.test series but not real soon.
This is starting to sound familiar.
I'm not really an expert on Linux SMP, but as I recall the coarse
locking used in the Linux net code was largely responsible for its poor
showing on the Mindcraft tests, in which they tested a quad processor
Intel box. There was some discussion at the time on linux-kernel about
rewriting the net code for 2.3. I don't know whether that happened. It
certainly hasn't been backported to 2.2.
By any chance is that 16MB transmitted with a single system call (i.e.
one large socket write)?
-- Jeff Sturm jeff.sturm@appnet.com-- To unsubscribe: send e-mail to axp-list-request@redhat.com with 'unsubscribe' as the subject. Do not send it to axp-list@redhat.com
This archive was generated by hypermail version 2a22 on Sat Jul 1 05:31:30 2000 PDT
Send any problems or questions about this archive to webmaster@alphalinux.org.