Performance issues in cups 1.3

[Johan Bernhardsson]

Okey did some testing during today.

Sent on 24 lp commands at the same time. On to almost identical machines. One with cups 1.3.3 and one with cups 1.2.7.

The one with 1.3.3 took almost 250 seconds before i got a response and the jobs ehere queued.

The one with 1.2.7 took about 3 seconds before i got a repsonse and the jobs where queued.

here are some screenshots fropm Glanceplus watching the cups pid. One with poll and one with select.

http://ryt.nu/select.png
http://ryt.nu/poll.png

Can anyone confirm this ? Using cups on hpux on parisc ?

/Johan "SmallOne" Bernhardsson



> The real problem is in the hp-ux clients from what ive seen. They use poll from what ive seen.
>
> Ill try to gather some strace output from the servers. And fetch the same from the clients.
>
> Ive tried to gather some debug2 logs. But i have to setup a test system for it. As we currently have to many printjobs to get any readable info.
>
> /Johan "SmallOne" Bernhardsson
>
>
> > Johan Bernhardsson wrote:
> > > ...
> > > When we push alot jobs through a client the load rises quite fast and we get delays.
> > >
> > > This seems to have started with the 1.3 release as we didnt have that in 1.2.7 that we ran before. The cpu usage is not the same.
> > >
> > > Has anyone else noted anything like this ?
> >
> > I haven't, but the scheduler in CUPS 1.3 no longer uses select() when
> > one of the alternate mechanisms is available - that would mean poll()
> > or epoll() on RHEL4.
> >
> > strace output from cupsd on the servers might be useful in seeing
> > where the scheduler is spending its time...
> >
> > --
> > ______________________________________________________________________
> > Michael R Sweet                        Senior Printing System Engineer
> >
>