Re: Solaris 10 not appearing to release memory in Update 2

From: J S (js.tech.mailer@gmail.com)
Date: Mon Jul 24 2006 - 18:37:51 EDT


Thanks to Xavier Mertens I found the problem - ZFS RAC Cache:

http://blog.rootshell.be/2006/07/08/the-arc-cache-effect/

23GB of kernel memory. Yum.

Page Summary Pages MB %Tot
------------ ---------------- ---------------- ----
Kernel 3017434 23573 36%
Anon 4265403 33323 51%
Exec and libs 30051 234 0%
Page cache 532497 4160 6%
Free (cachelist) 267142 2087 3%
Free (freelist) 250317 1955 3%

Total 8362844 65334
Physical 8227853 64280

cache buf buf buf memory alloc alloc
name size in use total in use succeed fail
------------------------- ------ ------ ------ --------- --------- -----
kmem_magazine_1 16 3470 65024 1048576 11898041 0
kmem_magazine_3 32 2546 14986 483328 951268 0
kmem_magazine_7 64 4492 11811 761856 903190 0
kmem_magazine_15 128 3687 7812 1015808 1941431 0
kmem_magazine_31 256 10203 10230 2703360 1127497 0
kmem_magazine_47 384 21172 21189 8265728 831649 0
kmem_magazine_63 512 69 270 147456 127331 0
kmem_magazine_95 768 1006 1010 827392 51023 0
kmem_magazine_143 1152 8152 8155 9543680 460634 0
kmem_slab_cache 56 281750 480385 27140096 11990662 0
kmem_bufctl_cache 24 541039 796989 19259392 12219580 0
kmem_bufctl_audit_cache 128 0 0 0 0 0
<...>
zio_buf_512 512 10285 32475 17735680 10837049 0
zio_buf_1024 1024 137 208 212992 7137075 0
zio_buf_1536 1536 23 45 73728 55786 0
zio_buf_2048 2048 9 36 73728 906783 0
zio_buf_2560 2560 5 36 98304 417873 0
zio_buf_3072 3072 10 48 147456 486991 0
zio_buf_3584 3584 1 54 196608 5160306 0
zio_buf_4096 4096 202 304 1245184 53667814 0
zio_buf_5120 5120 14 40 204800 220503 0
zio_buf_6144 6144 7 40 245760 261404 0
zio_buf_7168 7168 6 56 401408 706416 0
zio_buf_8192 8192 2 27 221184 4822143 0
zio_buf_10240 10240 5 20 204800 183073 0
zio_buf_12288 12288 16 38 466944 4178580 0
zio_buf_14336 14336 4 84 1204224 28184054 0
zio_buf_16384 16384 3823 3833 62799872 3397849 0
zio_buf_20480 20480 17 28 573440 1004540 0
zio_buf_24576 24576 8 28 688128 327889 0
zio_buf_28672 28672 2 14 401408 428699 0
zio_buf_32768 32768 5 17 557056 480454 0
zio_buf_40960 40960 12 20 819200 510290 0
zio_buf_49152 49152 8 18 884736 476010 0
zio_buf_57344 57344 7 14 802816 496411 0
zio_buf_65536 65536 495 504 33030144 522125 0
zio_buf_73728 73728 2 11 811008 207955 0
zio_buf_81920 81920 1 10 819200 183943 0
zio_buf_90112 90112 2 10 901120 170736 0
zio_buf_98304 98304 3 12 1179648 161862 0
zio_buf_106496 106496 2 10 1064960 141714 0
zio_buf_114688 114688 8 15 1720320 132246 0
zio_buf_122880 122880 4 10 1228800 112539 0
zio_buf_131072 131072 139861 139867 1152778240 105171646 0

On 7/24/06, J S <js.tech.mailer@gmail.com> wrote:
>
>
> I'm noticing an odd thing with 06/06 with the 118833-17 R&S patch set - I
> upgraded a v890 with 64GB RAM from 01/06 to 06/06, 4 memory hungry zones
> running oracle. The system commonly jumps to 62 or 63GB RAM usage, but
> doesn't seem to show itself letting go of a lot of the RAM (it will go down
> to 50 or 52GB), even when oracle's been brought down and all the zones have
> been dropped. prstat -a (or -z) is so bad about reporting the RAM in shared
> memory
> it's difficult to get a handle on what could be happening.
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers



This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:40:28 EDT