Re: Question about nmbclusters

[Peter Eisch <peisch%gmail.com@localhost>]

On 5/23/13 7:04 AM, "Greg Troxel" <gdt%ir.bbn.com@localhost> wrote:

>So, the 16384 upper limit may be there because it's not clearly safe to
>go higher, and it's rare to need more clusters than that.
>
>On a system with known memory and workload, certainly you can tweak
>settings.  I'd try NMBCLUSTERS=32768 and see how that goes.
>
>Also, it's possible that you are having a leak.  Does this happen more
>after long uptimes?  The following may be helpful:
>
>   vmstat -m|egrep 'Memory|Name|mcl'
>
>which will show the number of denied allocations.  Basically, watch
>"requests", "fail", and if requests-releases ends up higher and higher.

Greg, would you mind reading my tea leaves?  How should I read the mcl
stats from the following systems?

Host        Size Requests Fail Releases Pgreq Pgrel Npage Hiwat Minpg
Maxpg Idle
cottonmouth 2048     1095    0        0   552     0   552   552     4
131070   4
mouse       2048     3132    0        0  1570     0  1570  1570     4
131070   4
doily       2048   299331    0   298730 24798 24465   333   771     4
98304   11
cobra       2048     1137    0        0   573     0   573   573     4
98304    4
copperhead  2048     1587    0        0   798     0   798   798     4
98304    4
zebra       2048     1990    0        0   999     0   999   999     4
131070   4
buster      2048     3510    0     3116   607   393   214   294     4
8192   17
snoopy      2048     1205    0        0   607     0   607   607     4
131070   4
chowder     2048     3935    0        0  1972     0  1972  1972     4
8192    4
shark       2048      735    0        0   372     0   372   372     4
8192    4
horses      2048      736    0        0   372     0   372   372     4
8192    4
mivans      2048      735    0        0   372     0   372   372     4
8192    4
tbfw        2048     1026    0        0   517     0   517   517     4
16384    4
comfw       2048     1863    0      557   731    74   657   657     4
131070   4
ophfw       2048     2232    0        0  1120     0  1120  1120     4
8192    4
ophfw-old   2048     1204    0        0   606     0   606   606     4
16384    4
hoover      2048     5468    0     5134   447   267   180   220     4
131070  12
ivy         2048      555    0        0   262     0   282   282     4
65535    4
sara        2048     1685    0        0   847     0   847   847     4
8192    4
olivia      2048     5509    0     5196   541   367   174   223     4
8192   13
steph       2048     5327    0     4978   513   331   182   214     4
8192    7



I've struggled with tuning these various systems over the years.  If there
were a definitive guide for tuning stacks available it would be a
treasured resource.  I similarly battle with ipfilter tuning too but this
is probably out of scope in this thread.

At this point the systems above survive unless traffic through them
dramatically changes.  Maybe my fiddling with NMBCLUSTERS has only
coincidentally resolved the issues?

peter