Hi there,
Has anyone noticed very high size-32 allocations in Asterisk servers
with Digium hardware installed? Here is output from /proc/slabinfo:
size-32 23763586 23763586 32 119 1 : tunables 120
60 0 : slabdata 199694 199694 0
Here is the summary and first few rows from slabtop:
Active / Total Objects (% used) : 23850372 / 23890412 (99.8%)
Active / Total Slabs (% used) : 204139 / 204139 (100.0%)
Active / Total Caches (% used) : 95 / 134 (70.9%)
Active / Total Size (% used) : 756630.62K / 760089.77K (99.5%)
Minimum / Average / Maximum Object : 0.01K / 0.03K / 128.00K
OBJS ACTIVE USE OBJ SIZE SLABS OBJ/SLAB CACHE SIZE NAME
23764300 23764241 -80% 0.03K 199700 119 798800K size-32
5085 5085 100% 0.68K 1017 5 4068K ext3_inode_cache
51075 20557 40% 0.05K 681 75 2724K buffer_head
8008 3666 45% 0.27K 572 14 2288K radix_tree_node
9936 9863 99% 0.16K 432 23 1728K dentry_cache
8463 8463 100% 0.12K 273 31 1092K size-128
256 256 100% 3.00K 128 2 1024K biovec-(256)
As you can see, almost 800MB of memory on this box is taken up with
size-32 pages.
This particular server is a single CPU box running Asterisk 1.2.5 and
Zaptel 1.2.4 on RHEL4 and is a low-use, test box. Our two production
boxes are dual 3.4GHz Xeons running Asterisk 1.2.1 and Zaptel 1.2.1 on
RHEL4 SMP and exhibit the same issue (it was running into oom-killer
problems with low LOWMEM on one of them that triggered all of this).
Interestingly, we have an identical server to our test server that does
not have Asterisk or Zaptel installed, and it does not display this
issue.
Has anyone else encountered this issue? What does your slabtop look
like?
Any thoughts or ideas would be appreciated.
Regards,
--
Anthony Rodgers
Business Systems Analyst
District of North Vancouver
Web: http://www.dnv.org
RSS Feed: http://www.dnv.org/rss.asp