I've seen a problem in some machines running both linux
2.0.31pre2 and 2.0.29 with SCSI adapter Adaptec 2940, 3Com 3c590 and
3c900 cards. 64Megs of RAM, 100+ megs of swap, Pentiums. The problem is
that the computer seems to be facing a severe kernel memory leak. Both
have Triton II motherboards, IP alias and masquerading enabled. IPX is
enabled in both systems.
I don't know how I can really find out what happens, but the
symptoms I got are: virtual memory use increases steadily with time.
After some time (less than a day in both machines) the machine gives out
"Couldn't get a free page" message and then dies.
Shift-Scroll Lock gives increasing numbers in "Networking
buffers in use":
Jul 25 16:05:03 avantesma kernel: Networking buffers in use :
110468
Jul 25 16:05:03 avantesma kernel: Network buffers locked by drivers : 0
Jul 25 16:05:03 avantesma kernel: Total network buffer allocations :
4540263
Jul 25 16:05:03 avantesma kernel: Total failed network buffer allocs : 0
Jul 25 16:11:24 avantesma kernel: Networking buffers in use :
111586
Jul 25 16:11:24 avantesma kernel: Network buffers locked by drivers : 0
Jul 25 16:11:24 avantesma kernel: Total network buffer allocations :
4595171
Jul 25 16:11:24 avantesma kernel: Total failed network buffer allocs : 0
Jul 25 16:12:31 avantesma kernel: Networking buffers in use :
111874
Jul 25 16:12:31 avantesma kernel: Network buffers locked by drivers : 0
Jul 25 16:12:31 avantesma kernel: Total network buffer allocations :
4605792
Jul 25 16:12:31 avantesma kernel: Total failed network buffer allocs : 0
Jul 25 16:13:17 avantesma kernel: Networking buffers in use :
112210
Jul 25 16:13:17 avantesma kernel: Network buffers locked by drivers : 0
Jul 25 16:13:17 avantesma kernel: Total network buffer allocations :
4613933
Jul 25 16:13:17 avantesma kernel: Total failed network buffer allocs : 0
Jul 25 16:13:58 avantesma kernel: Networking buffers in use :
112241
Jul 25 16:13:58 avantesma kernel: Network buffers locked by drivers : 0
Jul 25 16:13:58 avantesma kernel: Total network buffer allocations :
4620366
Jul 25 16:13:58 avantesma kernel: Total failed network buffer allocs : 0
Jul 25 16:16:25 avantesma kernel: Networking buffers in use :
112690
Jul 25 16:16:25 avantesma kernel: Network buffers locked by drivers : 0
Jul 25 16:16:25 avantesma kernel: Total network buffer allocations :
4642504
Jul 25 16:16:25 avantesma kernel: Total failed network buffer allocs : 0
Jul 25 16:25:41 avantesma kernel: Networking buffers in use :
114234
Jul 25 16:25:41 avantesma kernel: Network buffers locked by drivers : 0
Jul 25 16:25:41 avantesma kernel: Total network buffer allocations :
4719646
Jul 25 16:25:41 avantesma kernel: Total failed network buffer allocs : 0
Not much is running, so there is no motive for the system to use
a lot of swap. A little after reboot, no swap is in use, but 8 hours
later, it uses about 20 MB of it. Load average increases when a process
is run, but much more than resonable (I think that's because of
trashing). When a shell prompt is inactive, it starts to take a long
time (say 1 or 2 seconds) to wake up when a key is pressed.
Am I right that those are because of trashing?
Anyone seeing that? That happened in a lot of different machines
with different kernel versions. Is that the 3c59x driver?
What should I try to find more about the problem?
Thanks in advance,
Andre
P.S. I only subscribe to the digest of this list.
-----
Andre Uratsuka Manoel Insite Internet
andre@insite.com.br +55-11-259-6979