rmilner@zia.aoc.nrao.edu (Ruth Milner) (11/14/90)
We have a Sun 3/260, running 4.0.3, which runs SLIP and has been running into
problems with streams allocation. Usually this just consists of SLIP making
"slip_output can't allocb" complaints, but today it spilled over into general
terminal usage. The root of the problem is that we are running out of the
larger streams buffers (excuse me if I haven't got the terminology correct;
I don't know very much about streams internals). The number of NBLK512/1024/
2048/4096 has already been increased once from the pathetically low values
they had originally, but it made no noticeable difference to SLIP.
Today we started seeing problems with the terminal driver being unable to
allocate streams buffers. "cat" was unable to output files over about 1K
directly to the screen (no problem with redirection to disk or any other
command). The output from "netstat -m" is as follows:
372/448 mbufs in use:
19 mbufs allocated to data
12 mbufs allocated to packet headers
103 mbufs allocated to socket structures
137 mbufs allocated to protocol control blocks
94 mbufs allocated to routing table entries
2 mbufs allocated to socket names and addresses
5 mbufs allocated to interface addresses
0/32 mapped pages in use
88 Kbytes allocated to network (52% in use)
0 requests for memory denied
streams allocation:
cumulative allocation
current maximum total failures
streams 22 23 157 0
queues 80 84 621 0
mblks 140 176 215604 0
total dblks 140 176 215604 90715
size 4 dblks 0 20 78588 0
size 16 dblks 3 22 32774 0
size 64 dblks 0 34 98291 0
size 128 dblks 52 55 2850 0
size 256 dblks 1 11 2089 0
size 512 dblks 28 28 267 11794
size 1024 dblks 28 28 689 45369
size 2048 dblks 28 28 56 33530
size 4096 dblks 0 0 0 22
As you can see, the problem is definitely in the streams area, not with
memory buffers or anything like that. Normally when this problem shows up,
the number of allocation failures is an order of magnitude lower than this.
So far all we have been able to do to cure this problem is reboot. Our
version of SLIP is up-to-date. Does anyone have any idea what's going on,
and/or how to fix it? It looks to me as though SLIP (or something, but it's
the likeliest possibility) is allocating buffers and not letting go of them
properly afterwards.
Thank you very much.
--
Ruth Milner
Systems Manager NRAO/VLA Socorro NM
rmilner@zia.aoc.nrao.edu