Did the ntop part of the hacking and it seems to work, except that it uses 99% of CPU 1 (well, kinda, since it's still Pentium 4 HT ). Concerning the glibc bug, it seems having been fixed in Mandrake 9.2 in a personal way (the patch is present but modified). Trying to replace syslog.c with the version proposed on the redhat site requires a lot of hack to compile glibc, but I did it. Except that it doesn't make it more stable, cause they replace pthread_cleanup_push/pop with some empty macros that have no effect. The result is that some services hang (things getting worse :) ). So I turned back to the original glibc version recompiled with linuxthreads_db enabled (disabled linuxthreads_db was a known bug in Mandrake 9.2). Having more time to run the SMP Mandrake kernel 2.4.22, I noticed however that it is less stable than the uniprocessor kernel (sometimes the system hangs anyway with or without ntop running), so I suppose Pentium 4 HT support is not great in Mandrake 9.2 overall. As ennoying it can be, I might consider upgrading the distribution before doing anything more. Anybody had similar troubles with these processors/dual channel DDR Intel mobo? Cause I have 2 identical machines and they behave both the same way. Florin On Fri, 2005-04-01 at 01:30, Burton Strauss wrote: > FWIW, here is the glibc bug: > http://sources.redhat.com/bugzilla/show_bug.cgi?id=671 > > Looks like it's supposed to be fixed in glibc >= 2.3, but I can't confirm > that. And it could have been backported into other versions by vendors. > > I'm beginning to think the safest thing is to just make the fix functional > on all platforms and live with the tiny deadlocks. Or make it a run time > switch. Preferences? > > -----Burton > > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf > Of Burton Strauss > Sent: Thursday, March 31, 2005 3:38 PM > To: [email protected] > Subject: RE: [Ntop-dev] SMP linux hangs with ntop > > self-LOCKED message is a red herring - read the back traffic. > > Again, this sounds like the syslog() not being thread safe problem (common > in kernel/glibc combinations prior to NPTL). SMP exposes it more often > (that's what I was able to find/fix it) and -t 4 also doesn't help. > > 1. Download the cvs. > 2. use utils/linuxrelease to figure out what configureextra file is being > used: > > $ utils/linuxrelease --distro > fedora > $ utils/linuxrelease --release > 2 > > Translates to configureextra/LINUXfedora2 > or configureextra/LINUXfedora > or configureextra/LINUX > > (1st match) > > Create a file for your version of Mandrake or edit the one that's already > there. Include the -DFORPRENPTL line from configureextra/LINUXredhat8.0 > > ./configure, make, make install as per usual. > > Post the info to ntop-dev for inclusion in the cvs. > > -----Burton > > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf > Of Florin > Sent: Thursday, March 31, 2005 1:15 PM > To: [email protected] > Subject: [Ntop-dev] SMP linux hangs with ntop > > I got ntop 3.1 running in daemon mode on a P4 HT 2.8GHz, dual-channel 512MB > DDRAM, 80GB SATA HDD, Mandrake Linux 9.2, kernel 2.4.22mdk, used as a > gigabit router. > When using the SMP kernel, accessing the ntop web server causes system hang > after the message: > "Mar 31 21:56:56 ns2 ntop[3560]: **WARNING** tryLockMutex() called > 'queuePacket' with a self-LOCKED mutex [EMAIL PROTECTED]:2219]" > Hard reset is the only way out. > If not running ntop, the system works fine with the SMP kernel. > The problem does not occur when booting the normal (uniprocessor) linux > kernel, even if the message is still there sometimes. > Is this a known problem with ntop? > > Florin > > > _______________________________________________ > Ntop-dev mailing list > [email protected] > http://listgateway.unipi.it/mailman/listinfo/ntop-dev
_______________________________________________ Ntop-dev mailing list [email protected] http://listgateway.unipi.it/mailman/listinfo/ntop-dev
