 | | | Source of shmget failure | Source of shmget failure 2005-10-03 - By Lachele Foley (Lists)
Back
I have a cluster of 63 HP Proliant DL360's (plus a DL380 serving as the root
node). Each has two processors and 2 G RAM.
# uname -a
Linux our.server 2.4.21-27.ELsmp #1 SMP Wed Dec 1 21:59:02 EST 2004 i686
i686 i386 GNU/Linux
One of my users routinely complains of shmget failures.
Usually, the reason is obvious. Sample reasons are that the entry in
/proc/sys/kernel/shmmax has "gone bad" (I've fixed this, I think) or that he
has overrun the available hard drive space. In the latter case, the machine
reboots itself and whatever reports errors to him calls that a shmget
failure.
My best guess about the last failure is that there wasn't enough available
RAM. free -o on the offending node gave:
# free -o
total used free shared buffers
cached
Mem: 2055492 834620 1220872 0 36436
734132
Swap: 2097112 2096720 392
The trouble with this picture is that the node wasn't running anything
except the usual background noise -- it certainly should not have required
0.83 G to do it. And, the machine had been in this "idle" state for a good
12+ hours. Shouldn't that be ample time for unused memory to be returned to
"free" state?
After I rebooted, there wasn't any significant change in the number and type
of running processes. However, there was a *lot* more free memory.
I started watching memory use. In the course of an hour, free reported that
the memory use steadily increased (from about 256000 to 260500 -- it makes a
lovely graph).
So, I've got my nose poked into "Understanding Virtual Memory" at this
location:
http://www.redhat.com/magazine/001nov04/features/vm/
Am I looking in the right place? If not, I will gladly accept any
suggestions. If I am in the right place, well, I'll still gladly accept
suggestions, but don't feel obligated.
Thanks!
:-) Lachele
I have a cluster of 63 HP Proliant DL360's (plus a DL380 serving as the root
node). Each has two processors and 2 G RAM.<br>
<br>
# uname -a<br>
Linux our.server 2.4.21-27.ELsmp #1 SMP Wed Dec 1 21:59:02 EST 2004 i686 i686
i386 GNU/Linux<br>
<br>
One of my users routinely complains of shmget failures.<br>
<br>
Usually, the reason is obvious. Sample reasons are that the entry
in /proc/sys/kernel/shmmax has "gone bad" (I've fixed this, I think)
or
that he has overrun the available hard drive space. In the latter
case, the machine reboots itself and whatever reports errors to him
calls that a shmget failure.<br>
<br>
My best guess about the last failure is that there wasn't enough available RAM.
free -o on the offending node gave:<br>
<br>
<pre># free -o<br>
total used
free shared  
;buffers<br> cached<br>Mem:
2055492 834620 1220872
0  
;36436<br> 734132<br>Swap:  
;2097112 2096720 392
<br>
<br></pre>
<div style="text-align: left;"><span style="font-family: arial,sans-serif;"><
/span>The
trouble with this picture is that the node wasn't running anything
except the usual background noise -- it certainly should not have
required 0.83 G to do it. And, the machine had been in this
"idle" state for a good 12+ hours. Shouldn't that be ample time
for unused memory to be returned to "free" state?<br>
<br>
After I rebooted, there wasn't any significant change in the number and
type of running processes. However, there was a *lot* more free
memory.<br>
<br>
I started watching memory use. In the course of an hour, free
reported that the memory use steadily increased (from about 256000 to
260500 -- it makes a lovely graph). <br>
<br>
So, I've got my nose poked into "Understanding Virtual Memory" at
this location:<br>
<br>
<a href="http://www.redhat.com/magazine/001nov04/features/vm/">http://www
.redhat.com/magazine/001nov04/features/vm/</a><br>
<br>
Am I looking in the right place? If not, I will gladly accept any
suggestions. If I am in the right place, well, I'll still gladly
accept suggestions, but don't feel obligated.<br>
<br>
Thanks!<br>
<br>
:-) Lachele<br>
<br>
</div>
--
Taroon-list mailing list
Taroon-list@(protected)
https://www.redhat.com/mailman/listinfo/taroon-list
|
|
|