On Fri, Feb 18, 2000 at 12:50:02AM -0800, Gordon Messmer wrote:
> Hal Burgiss wrote:
> > Top seems to show X CPU % to fluctuate a fair amount. Occasional
> > spikes from less than 1% to 20-30% while I am just sitting watching
> > top. If I constantly drag a window around it will bump up to 80% or
> > so. On the otherhand ps almost always shows X at about 4%. Little to
> > no fluctuation -- no matter what. Who's right?
>
> It appears to be "not a bug". I brought this up on redhat's bug
> tracking system a couple of months ago, and Alan (himself) seems to
> think that some changes made to the UNIX socket handling make
> applications using UNIX sockets (that's all of your X stuff)
> _appear_ to take up more processor time than they actually do. I
> believe they are being syncronised with the kernel's load testing.
> Since operations on UNIX sockets get sync'ed with the test, they get
> reported more often than they probably should. As far as I can
> tell, however, the performance of UNIX sockets (or anything else)
> doesn't suffer from this behavior, even though I went in convinced
> that they did.
>
> Unless you notice things slowing down where I did not (and I wrote a
> small app to test it), then don't worry about it.
I haven't noticed a slowdown, but would like to have an idea what kind
of load the system is under. The other thing I didn't mention, is that
I flashed a new BIOS, and these spikes became constant loads. Cold
boot into X was taking 30% CPU with 2.2.15pre7. I backed down to
2.2.14 and the constant load is gone, but still have the spikes.
Curious too that some apps are not effected at all -- Netscape and
gvim so no bump at all. Some are marginally effected -- Eterm, wmaker,
asMixer. Strange...
Also I see this, especially during the spikes:
74 processes: 71 sleeping, 3 running, 0 zombie, 0 stopped
Are 3 concurrent processes possible with 2 CPUs? I didn't think so,
and don't see how.
The more I learn, the more confused I get ;)
--
Hal B
[EMAIL PROTECTED]
--
--
To unsubscribe: mail [EMAIL PROTECTED] with "unsubscribe"
as the Subject.
