Re: ps performance sucks (was Re: dcache_rcu [performance results])

[Erik Andersen <andersen@codepoet.org>]

On Mon Nov 04, 2002 at 09:44:07PM -0800, Martin J. Bligh wrote:
> > Hehe.  You just reinvented my old /dev/ps driver.  :)
> 
> Indeed, sounds much more like a /dev thing than a /proc thing
> at this point ;-)
> 
> > http://www.busybox.net/cgi-bin/cvsweb/busybox/examples/kernel-patches/devps.patch.9_25_2000?rev=1.2&content-type=text/vnd.viewcvs-markup
> > 
> > This is what Linus has to say on the subject:
> > 
> > ... If the problem is that /proc
> >     is too large, then the right solution is to just clean up
> >     /proc.  Which is getting done.  And yes, /proc will be larger
> >     than /dev/ps, but I still find that preferable to having two
> >     incompatible ways to do the same thing.
> 
> Ummm ... how do we make /proc smaller than 1 file to open per PID?
> It's pretty easy to get it down that far. But it still sucks.
> 
> >     I do dislike /dev/ps mightily.
> 
> Well it can't be any worse than the current crap. At least it'd 
> stand a chance in hell of scaling a little bit. So I took a very 
> quick look ... what syscalls are you reduced to per pid, one ioctl 
> and one read?

As I implemented it, it was one ioctl per pid...  Of course 
it could be easily modified to be one syscall, one read from
the /dev/ps char device, or similar...

 -Erik

--
Erik B. Andersen             http://codepoet-consulting.com/
--This message was written using 73% post-consumer electrons--