Problem: Alpha, 2.2.7 "fork: Cannot allocate memory"

furio ercolessi (furio@spin.it)
Sun, 20 Jun 1999 16:37:52 METDST


Hello linux-kernel,

I am not a list reader but I have been recommended to let you guys
aware of my problem.

h/w: Alpha 21164/533 MHz, LX board, 512 MB RAM
base installation: RedHat 5.2
kernel: 2.2.7 compiled with: gcc version egcs-2.90.29 980515 (egcs-1.0.3 release)
swap: 256MB in a single partition
defaults changed:
NR_OPEN from 1024 to 16384 (fs.h, limits.h)
NR_FILE from 4096 to 16384 (fs.h)
OPEN_MAX from 256 to 4096 (limits.h)
__FD_SETSIZE from 1024 to 16384 (posix_types.h)
operational conditions: postfix (MTA) and squid (web caching)
in a production environment

Problem: occasionally the system refuses to start new processes,
apparently without a valid reason. Here is an example:
-------------------------------------------------
# man ps
Formatting page, please wait...
sh: fork: Cannot allocate memory
sh: fork: Cannot allocate memory
Error executing formatting or display command.
[...]
# free
total used free shared buffers cached
Mem: 516720 510696 6024 71416 214536 97896
-/+ buffers/cache: 198264 318456
Swap: 262136 2280 259856
-------------------------------------------------

Of the 198MB used, about 120MB were used by squid.
There were about 100 processes active.
NR_TASKS was the default of 512; raising it to 4096 did not help.
I also tried to add an additional 1GB of swap (on a file), which
did not help either.

This problem did not appear when running postfix only (which
generates quite a bit of processes but does not use much
memory), or squid only (which uses a lot of memory but is
essentially a single-process application): it required both
products to operate simultaneously to manifest.
Memory allocation by squid appeared to be regular (that is,
"Total in use" only slightly larger than "Total accounted").

I moved to 2.0.37 and the problem disappeared. Unfortunately
I am not in the position to do further testing with 2.2.x right now.

furio ercolessi
Spin - Trieste (Italy)

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/