Under some heavy load circumstances, it seems that smbd can lock up and consume 100% cpu.
This happens when the Heimdal library can't find a /dev/random and start using itimer/SIGALRM to get entropy from the system scheduler. It appears that the SIGALRM signal is getting lost (possibly from mixed usage with other timers that the samba server uses) and the process simply consumes CPU without servicing any more network requests. It has to be killed manually.
Our advice is to ensure that your operating system has a /dev/random. This is not always available, and sometimes must be installed as an OS patch.
HPUX 11.11 users can install KRNG11 from http://software.hp.com/portal/swdepot/displayProductInfo.do?productNumber=KRNG11I
HPUX 11.00 users can visit http://www.josvisser.nl/hpux11-random
Solaris 8 users should install the following OS patch from sunsolve.sun.com
- solaris8/sparc: OS patch 112438
- solaris8/x86: OS patch 112439
Entropy gathering daemons do not help, unless you make /dev/random a named pipe or similar.
This is logged as bug 256.