[JDEV] User logins slow down?

Tue Sep 4 16:16:24 CDT 2001

We haven't tried it with jpoll or (forget the name of the other one).  We
haven't been successful at getting either to work well enough to warrant
testing :)

Dustin, maybe you could check into jpolld again...

> -----Original Message-----
> From: temas [mailto:temas at box5.net]
> Sent: Monday, September 03, 2001 3:24 PM
> To: jdev at jabber.org
> Subject: Re: [JDEV] User logins slow down?
> 
> 
> Ok, my though would be that it's actually c2s.  c2s is 
> completely suboptimal
> because it has to traverse the select list to find which item 
> has data.  Now
> if we're getting a lot of events firing or a stampeding herd 
> of connecting
> sockets this could cause a _ton_ of list walks and possibly cause the
> apparent slow down.. Have you tried this with jpoll or 
> something else that
> would better handle connections?
> 
> --temas
> 
> ----- Original Message -----
> From: "Dustin Puryear" <dpuryear at usa.net>
> To: <jdev at jabber.org>
> Sent: Friday, August 31, 2001 12:54 PM
> Subject: RE: [JDEV] User logins slow down?
> 
> 
> > On Fri, 2001-08-31 at 12:14, Colin Madere wrote:
> > >
> > > If it's the one you're running on AIX, I know Matt 
> checked out latest
> from
> > > CVS before he made any mods to get it running on AIX... 
> sooo.. it's
> > > 1.4.<latest from CVS>
> > >
> > > But you've seen this on the official 1.4.1 release on Linux, also.
> Correct
> > > Dustin?
> >
> > Yes. Below is every 10'nth login time.
> >
> > Linux:
> >
> > [dpuryear at crack testsuite]$ ./pasvlogin -h crack -u 100 | 
> awk '$1 % 10
> > == 0 {print}'
> > 0     999280305.494653 999280305.508374 0.013721
> > 10    999280305.686730 999280305.701709 0.014979
> > 20    999280305.886727 999280305.904158 0.017431
> > 30    999280306.286775 999280306.333869 0.047094
> > 40    999280306.856944 999280306.911021 0.054077
> > 50    999280307.506850 999280307.575894 0.069044
> > 60    999280308.306917 999280308.388676 0.081759
> > 70    999280309.246955 999280309.338412 0.091457
> > 80    999280310.267189 999280310.375869 0.108680
> > 90    999280311.477229 999280311.593550 0.116321
> >
> > AIX:
> >
> > [dpuryear at crack testsuite]$ ./pasvlogin -h 
> viagra.vedalabs.com -u 100 |
> > awk '$1 % 10 == 0 {print}'
> > 0     999280337.245336 999280337.262162 0.016826
> > 10    999280337.408921 999280337.419975 0.011054
> > 20    999280337.608931 999280337.625213 0.016282
> > 30    999280337.858746 999280337.880329 0.021583
> > 40    999280338.158831 999280338.187947 0.029116
> > 50    999280338.528916 999280338.561017 0.032101
> > 60    999280338.948951 999280338.991652 0.042701
> > 70    999280339.448982 999280339.495653 0.046671
> > 80    999280340.009029 999280340.062667 0.053638
> > 90    999280340.641434 999280340.700998 0.059564
> >
> > Almost linear growth.
> >
> > Regards, Dustin
> >
> > >
> > > Colin
> > >
> > > > -----Original Message-----
> > > > From: Dustin Puryear
> > > > Sent: Friday, August 31, 2001 11:32 AM
> > > > To: jdev at jabber.org
> > > > Subject: Re: [JDEV] User logins slow down?
> > > >
> > > >
> > > > On Thu, 2001-08-30 at 20:43, temas wrote:
> > > > > >and AIX it grows from .1 seconds to .7 seconds as n 
> = 1 -> 1000.
> > > > > >
> > > > > >Is that expected? Should login time increase as the number
> > > > of online
> > > > > >users increases?
> > > > > >
> > > > > >Regards, Dustin
> > > > > >
> > > > > Hrm... I'll have to look at some data structures, but I
> > > > don't think it
> > > > > would.  Unless is was traversing some list, but that
> > > > doesn't make sense
> > > > > for login.  I'm assuming this is jabberd 1.4.1?
> > > >
> > > > That's what I was saying when I discussed this over 
> here. I don't see
> > > > why additional logins would slow down jabberd, but it 
> appears to be
> > > > doing just that.
> > > >
> > > > I performed a second test, just to confirm. I had a 
> process logging in
> > > > 1000 users, and I noticed login time went from .1 to .7
> > > > seconds as n1 =
> > > > 1 -> 1000. Ok, so I started another process logging in 
> users n2 = 1001
> > > > -> 2000. As n2 logged in it took the same amount of 
> time as n1 was
> > > > taking at that time, and when I killed the first 
> process (n1), n2's
> > > > login times decreased dramatically afterward, although 
> there seemed to
> > > > be a initial pause as jabberd dropped the n1 users.
> > > >
> > > > Regarding the version, I'm looking in the README and 
> see Jabber 1.4.
> > > > Matt, is that correct?
> > > >
> > > > Regards, Dustin
> > > >
> > > > >
> > > > > --temas
> > > > >
> > > > >
> > > > >
> > > > > _______________________________________________
> > > > > jdev mailing list
> > > > > jdev at jabber.org
> > > > > http://mailman.jabber.org/listinfo/jdev
> > > > >
> > > > --
> > > > Dustin Puryear <dpuryear at usa.net>
> > > > http://members.telocity.com/~dpuryear
> > > > In the beginning the Universe was created.
> > > > This has been widely regarded as a bad move. - Douglas Adams
> > > >
> > > > _______________________________________________
> > > > jdev mailing list
> > > > jdev at jabber.org
> > > > http://mailman.jabber.org/listinfo/jdev
> > > >
> > > _______________________________________________
> > > jdev mailing list
> > > jdev at jabber.org
> > > http://mailman.jabber.org/listinfo/jdev
> > >
> > --
> > Dustin Puryear <dpuryear at usa.net>
> > http://members.telocity.com/~dpuryear
> > In the beginning the Universe was created.
> > This has been widely regarded as a bad move. - Douglas Adams
> >
> > _______________________________________________
> > jdev mailing list
> > jdev at jabber.org
> > http://mailman.jabber.org/listinfo/jdev
> >
> 
> _______________________________________________
> jdev mailing list
> jdev at jabber.org
> http://mailman.jabber.org/listinfo/jdev
> 