On Thu, 18 Dec 2003, [gb2312] Xin LI/ÀîöÎ wrote: > One of my friend is operating servers on a big ICP in China. They found that > if they run FreeBSD 4.9 or FreeBSD 5.1-RELEASE on their server, which has > more than 4GB RAM installed, then they will occasionaly have the server > panic'ed, or even rebooted without any reason. > > The server hardware is: P4-Xeon-2.4G*2, 4G RAM(DDR266), 146G*2 on asr(4) > RAID-1. The kernel is not quite different from GENERIC, except that we have > AUTO_EOI1 enabled. When PAE is enabled, then the kernel will not recognize > the asr(4) device, and thus would refuse to boot from it, so we have it > disabled. PAE only has effect for >4GB (strictly), so exactly 4GB will work fine with !PAE. > This problem afffects 4.9-RELEASE and 5.1-RELEASE. As the ICP is a > conservative user, they will not run -STABLE, nor -CURRENT. I have turned > the debugging symbols on their server and grabbed the following backtraces. > Also, I have found how to reproduce this problem on a machine running with > > 4GB memory without a PAE configuration, which, in my opinion, should not > cause a kernel panic because it's just a resource hungry program and all it > will do should, at most, hang the system, or even being killed directly by > the operating system. Its probably due to poor tuning in this case. See the tuning(7) man page and innumerable mailing lists posts. Basically, you want to turn down maxusers (128 or so works best) and maybe change the kernel/user boundary and kmem sizes. vmstat -m is useful to keep track of kernel memory use. Also forkbombs (which is what your program is, effectively) are best controlled by using user resource limits. They're there for a reason. See the login.conf(5) man page. > panic: pmap_new_proc: u_map allocation failed kmem exhaustion condition in -stable. -- Doug White | FreeBSD: The Power to Serve dwhite_at_gumbysoft.com | www.FreeBSD.orgReceived on Wed Dec 17 2003 - 15:58:14 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:34 UTC