IDS Forum

Re: I/O writes

Posted By: Art Kagel
Date: Friday, 17 July 2009, at 10:39 a.m.

In Response To: I/O writes (KAMRAN HAQ)

Kamran,

You are running IDS 11.50 which implements non-blocking checkpoints. In
this and all following releases CHUNK writes are actually preferred. Since
your checkpoints are non-blocking except for a very brief setup phase, you
should not need to worry about long checkpoints any longer. Indeed, you are
more concerned with the blocking time displayed at the end of the checkpoint
trace in the message log. If you are not sure how to read that, then post
some message log output that shows the specific checkpoints that are giving
yoiu concern.

In general, and this was always true, CHUNK writes are more efficient for
the OS and IO subsytems and even for the Informix engine itself than LRU
writes. The reason that a balance of about 60-70% LRU writes to 30-40%
CHUNK writes was recommended in earlier releases for OLTP systems was that
extensive chunk writes can result in longer checkpoints and prior to IDS v11
checkpoints blocked transactions and user sessions in critical code sections
could delay the start of a checkpoint which would increase the blockout time
for other sessions. With non-blocking checkpoints, this is no longer a
problem so IBM and the user community have reverted back to recommending
configuring higher LRU_MIN_DIRTY & LRU_MAX_DIRTY levels to shift the balance
to mostly CHUNK writes or even 100% CHUNK writes.

Art

Art S. Kagel
Oninit (www.oninit.com)
IIUG Board of Directors (art@iiug.org)

Disclaimer: Please keep in mind that my own opinions are my own opinions and
do not reflect on my employer, Oninit, the IIUG, nor any other organization
with which I am associated either explicitly or implicitly. Neither do
those opinions reflect those of other individuals affiliated with any entity
with which I am affiliated nor those of the entities themselves.

On Fri, Jul 17, 2009 at 9:33 AM, KAMRAN HAQ <khaq@i2cinc.com> wrote:

> Hi,
> Below are the stats gathered from an IDS instance (restarted around 2 hours
> ago). We see that checkpoints are taking too much time sometimes. Problems
> looks in I/O. How can we improve "LRU writes" and reduce "Chunk writes" as
> I
> read that "LRU writes" are better. Also how to keep max length (onstat -g
> ioq)
> shorter.
> OS: SunOS 5.10 sun4v sparc
> regards,
> Kamran
> --------------------------------------------------------------
>
> $ onstat -F
>
> IBM Informix Dynamic Server Version 11.50.FC3W1 -- On-Line (Prim) -- Up
> 01:09:50 -- 921600 Kbytes
>
> Fg Writes LRU Writes Chunk Writes
> 0 0 622567
>
> address flusher state data # LRU Chunk Wakeups Idle Tim
> 119c6d870 0 I 0 0 32 4202 4172.668
> 119c6e0b8 1 I 0 0 36 4190 4154.733
> 119c6e900 2 I 0 0 28 4197 4173.793
> 119c6f148 3 I 0 0 30 4196 4166.957
> 119c6f990 4 I 0 0 30 4194 4163.863
> 119c701d8 5 I 0 0 34 4179 4145.377
> 119c70a20 6 I 0 0 36 4181 4145.399
> 119c71268 7 I 0 0 29 4197 4167.696
> 119c71ab0 8 I 0 0 28 4197 4169.929
> 119c722f8 9 I 0 0 24 4199 4176.918
> 119c72b40 10 I 0 0 24 4196 4172.204
> 119c73388 11 I 0 0 23 4192 4171.186
> 119c73bd0 12 I 0 0 22 4192 4171.252
> 119c74418 13 I 0 0 20 4196 4176.573
> 119c74c60 14 I 0 0 20 4193 4175.085
> 119c754a8 15 I 0 0 20 4192 4174.220
> 119c75cf0 16 I 0 0 20 4188 4170.391
> 119c76538 17 I 0 0 20 4193 4173.006
> 119c76d80 18 I 0 0 19 4188 4170.294
> 119c775c8 19 I 0 0 18 4185 4166.483
> 119c77e10 20 I 0 0 17 4175 4157.900
> 119c78658 21 I 0 0 17 4153 4135.801
> 119c78ea0 22 I 0 0 16 4147 4130.073
> 119c796e8 23 I 0 0 16 4148 4131.168
>
> states: Exit Idle Chunk Lru
>
> $ onstat -g ioq
>
> IBM Informix Dynamic Server Version 11.50.FC3W1 -- On-Line (Prim) -- Up
> 01:10:04 -- 921600 Kbytes
>
> AIO I/O queues:
> q name/id len maxlen totalops dskread dskwrite dskcopy
> drda_dbg 0 0 0 0 0 0 0
> sqli_dbg 0 0 0 0 0 0 0
> adt 0 0 0 0 0 0 0
> msc 0 0 2 10356 0 0 0
> aio 0 0 34 4957068 26 4954270 0
> pio 0 0 1 4208 0 4208 0
> lio 0 0 1 934928 0 934928 0
> gfd 3 0 3 44271 44168 103 0
> gfd 4 0 5 183 52 131 0
> gfd 5 0 31 164375 163919 456 0
> gfd 6 0 16 5707 1454 4253 0
> gfd 7 0 2 25179 25179 0 0
> gfd 8 0 31 657803 657480 323 0
> gfd 9 0 14 3977 3879 98 0
> gfd 10 0 31 581371 577245 4126 0
> gfd 11 0 30 46808 46730 78 0
> gfd 12 0 11 3140 3076 64 0
> gfd 13 0 1 1153 1153 0 0
> gfd 14 0 45 245420 244784 636 0
> gfd 15 0 29 15783 15335 448 0
> gfd 16 0 1 17 17 0 0
> gfd 17 0 1 20 20 0 0
> gfd 18 0 30 224616 224114 502 0
> gfd 19 0 2 3273 3266 7 0
> gfd 20 0 30 62404 61709 695 0
> gfd 21 0 30 170825 169045 1780 0
> gfd 22 0 30 11490 11065 425 0
> gfd 23 0 31 232938 231740 1198 0
> gfd 24 0 30 68956 65598 3358 0
> gfd 25 0 31 175224 174670 554 0
> gfd 26 0 31 253559 251803 1756 0
> gfd 27 0 30 201740 199354 2386 0
> gfd 28 0 43 717590 707646 9944 0
> gfd 29 0 31 790372 774676 15696 0
> gfd 30 0 1 1 1 0 0
> gfd 31 0 46 719426 679486 39940 0
> gfd 32 0 32 853121 759948 93173 0
> gfd 33 0 48 1158744 1006962 151782 0
> gfd 34 0 1 83 83 0 0
> gfd 35 0 41 728786 654252 74534 0
> gfd 36 0 32 692805 615643 77162 0
> gfd 37 0 31 1075115 934510 140605 0
> gfd 38 0 1 1 1 0 0
> gfd 39 0 0 0 0 0 0
> gfd 40 0 0 0 0 0 0
> gfd 41 0 0 0 0 0 0
> gfd 42 0 0 0 0 0 0
> gfd 43 0 0 0 0 0 0
> gfd 44 0 0 0 0 0 0
> gfd 45 0 0 0 0 0 0
> gfd 46 0 0 0 0 0 0
> gfd 47 0 0 0 0 0 0
> gfd 48 0 0 0 0 0 0
> gfd 49 0 0 0 0 0 0
> gfd 50 0 0 0 0 0 0
> gfd 51 0 0 0 0 0 0
> gfd 52 0 0 0 0 0 0
> gfd 53 0 0 0 0 0 0
> gfd 54 0 0 0 0 0 0
> gfd 55 0 0 0 0 0 0
> gfd 56 0 0 0 0 0 0
> gfd 57 0 0 0 0 0 0
> gfd 58 0 0 0 0 0 0
> gfd 59 0 0 0 0 0 0
> gfd 60 0 0 0 0 0 0
> gfd 61 0 0 0 0 0 0
> gfd 62 0 0 0 0 0 0
> gfd 63 0 0 0 0 0 0
> gfd 64 0 0 0 0 0 0
> gfd 65 0 0 0 0 0 0
> gfd 66 0 0 0 0 0 0
> gfd 67 0 0 0 0 0 0
> gfd 68 0 0 0 0 0 0
> gfd 69 0 0 0 0 0 0
> gfd 70 0 0 0 0 0 0
> gfd 71 0 0 0 0 0 0
> gfd 72 0 0 0 0 0 0
> gfd 73 0 0 0 0 0 0
> gfd 74 0 0 0 0 0 0
>
> $ onstat -p
>
> IBM Informix Dynamic Server Version 11.50.FC3W1 -- On-Line (Prim) -- Up
> 01:10:19 -- 921600 Kbytes
>
> Profile
> dskreads pagreads bufreads %cached dskwrits pagwrits bufwrits %cached
> 9355610 16947434 166184193 94.37 1568001 2150214 6486797 75.88
>
> isamtot open start read write rewrite delete commit rollbk
> 173486072 3649325 17192634 82339872 730432 30257 463299 941887 35
>
> gp_read gp_write gp_rewrt gp_del gp_alloc gp_free gp_curs
> 0 0 0 0 0 0 0
>
> ovlock ovuserthread ovbuff usercpu syscpu numckpts flushes
> 0 0 0 12088.14 2358.88 27 30
>
> bufwaits lokwaits lockreqs deadlks dltouts ckpwaits compress seqscans
> 550159 79 123453819 0 0 67 150460 120453
>
> ixda-RA idx-RA da-RA RA-pgsused lchwaits
> 2025249 1361 1499088 3519878 468120
>
>
>
> *******************************************************************************
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
>

--001636c5a9bf9d7014046ee7c1d3

Messages In This Thread

• I/O writes
  KAMRAN HAQ -- Friday, 17 July 2009, at 9:33 a.m.
  • Re: I/O writes
    Art Kagel -- Friday, 17 July 2009, at 10:39 a.m.
    • Re: I/O writes
      KAMRAN HAQ -- Friday, 17 July 2009, at 11:00 a.m.
      • Re: I/O writes
        KAMRAN HAQ -- Friday, 17 July 2009, at 11:06 a.m.
        • Re: I/O writes
          Art Kagel -- Friday, 17 July 2009, at 11:15 a.m.
      • Re: I/O writes
        Art Kagel -- Friday, 17 July 2009, at 11:11 a.m.
        • Re: I/O writes
          KAMRAN HAQ -- Friday, 17 July 2009, at 11:25 a.m.
          • Re: I/O writes
            Art Kagel -- Friday, 17 July 2009, at 11:36 a.m.

IDS Forum is maintained by Administrator with WebBBS 5.12.