[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

some in-kernel inode problems






  Hi,



   We are running one of our major proxies here on a DELL PowerEdge 2200,
256MB ECC SDRAM, 3x4GB UW SCSI disks, Adaptec UW controller, using a
rather modified (by me) version of RedHat 4.2 (I didn't want glibc).
(by modified I mean stable libc - 5.4.38, stable binutils, stable Net-Kit,
everything updated, and not the old stuff that is default for RedHat 4.2)

   I changed the kernel to allow 1024 fds, (2.0.33), and I am using squid
1.1.20. THe cache size is about 10GB, total swap is about 390MB. It's
using a DEC Tulip based 10/100 ethernet card, running at fd 100Mbps.

I have a rather annoying problem with it, after it is up for say, 1-2
weeks at a time. (I forgot to mention, squid is shutdown/restarted every
day from cron, to reduce the memory usage, since after 24hrs it takes
almost all real memory, and squid sucks if using swap). Anyway, the
problem is that I get these messages:

Apr 17 15:25:30 wwwproxy1 kernel: VFS: No free inodes - contact Linus
Apr 17 15:25:58 wwwproxy1 kernel: VFS: No free inodes - contact Linus
Apr 17 15:27:58 wwwproxy1 kernel: VFS: No free inodes - contact Linus
Apr 17 15:29:58 wwwproxy1 kernel: VFS: No free inodes - contact Linus
Apr 17 15:31:53 wwwproxy1 kernel: VFS: No free inodes - contact Linus

After this, the machine remains alive, inetd services respond, but get
hung after inetd spawns them, sshd dies, squid dies. Unless one reboots,
the machine remains in this pitiful state.

>From asking around my friends on #unix on EFnet, I gathered that
in-kernel inodes should never be deplenished. 

I need to find a solution for this. We are benchmarking two servers, this
one against an identical server, only running SlowLaris 2.6 (x86). The
outcome of the benchmark, (stability, reliability) will determine what OS
will be used for PC based Unix servers at Tel-Aviv University. I see this
as an important test for Linux.

I would like to know if anyone knows what the problem is, how to fix it,
or other suggestions.

Best,

--Ariel

   +---------------------------------------------------------------+
   | Ariel Biener                                                  |
   | e-mail: ariel@post.tau.ac.il        Work ph: 03-6406086       |
   | fingerprint = 07 D1 E5 3E EF 6D E5 82 0B E9 21 D4 3C 7D 8B BC |
   +---------------------------------------------------------------+