aix 5.1(64bit). rs6000(h80/4way) 8gig of memory

We are running a Progress(32bit) database application. We lowered MinPerm
and Maxperm since we turned on directio within the db startup. At curtain
times during the day using Topas we see what appears to be disk waits.

What commands or scripts could I use to track this down.


I had are systems person run (vmtune -a) during that period:

total mempools = 1
total_frame_sets = 2
vmm_page_color_on = 0
memory frames = 2097136
maxpin = 1677709
minperm = 200615
maxperm = 401231
maxclient = 401231
numperm = 1268156
numclient = 394719
numcompress = 0
maxpin % = 80.0
minperm % = 10.0
maxperm % = 20.0
pgs_thresh % = 0.0
maxclient % = 20.0
numperm % = 63.2
numclient % = 19.7
numcompress % = 0.0
minpgahead = 2
maxpgahead = 8
minfree = 3840
maxfree = 4352
pd_npages = 65536
maxrandwrt = 32
numclust = 1
jfs_clread_enabled = 0
jfs_use_read_lock = 1
npswarn = 163072
npskill = 40768
numfsbufs = 2048
hd_pbuf_cnt = 704
lvm_bufcnt = 9
lrubucket = 131072
defps = 1
sync_release_ilock = 0
pta_balance_thresh = 50.000000
nokilluid = 0
v_pinshm = 0
strict_maxperm = 0
j2_nPagesPerWriteBehindCluster = 32
j2_maxRandomWrite = 0
j2_nRandomCluster = 0
j2_minPageReadAhead = 2
j2_maxPageReadAhead = 8
j2_nBufferPerPagerDevice = 512
lgpg_regions = 0
lgpg_size = 0
mem_affinity_on = 0
num_spec_dataseg = 0
spec_dataseg_int = 512
lru_poll_interval =
hd_pendqblked = 380
psbufwaitcnt = 18030
fsbufwaitcnt = 13497
rfsbufwaitcnt = 0
xpagerbufwaitcnt = 7001

Does anything listed here appear to be causing this.

Thanks.

Hoosbruin wrote:
> aix 5.1(64bit). rs6000(h80/4way) 8gig of memory
>
> We are running a Progress(32bit) database application. We lowered MinPerm
> and Maxperm since we turned on directio within the db startup. At curtain
> times during the day using Topas we see what appears to be disk waits.
>
> What commands or scripts could I use to track this down.
>
>
> I had are systems person run (vmtune -a) during that period:
>
> total mempools = 1
> total_frame_sets = 2
> vmm_page_color_on = 0
> memory frames = 2097136
> maxpin = 1677709
> minperm = 200615
> maxperm = 401231
> maxclient = 401231
> numperm = 1268156
> numclient = 394719
> numcompress = 0
> maxpin % = 80.0
> minperm % = 10.0
> maxperm % = 20.0
> pgs_thresh % = 0.0
> maxclient % = 20.0
> numperm % = 63.2
> numclient % = 19.7
> numcompress % = 0.0
> minpgahead = 2
> maxpgahead = 8
> minfree = 3840
> maxfree = 4352
> pd_npages = 65536
> maxrandwrt = 32
> numclust = 1
> jfs_clread_enabled = 0
> jfs_use_read_lock = 1
> npswarn = 163072
> npskill = 40768
> numfsbufs = 2048
> hd_pbuf_cnt = 704
> lvm_bufcnt = 9
> lrubucket = 131072
> defps = 1
> sync_release_ilock = 0
> pta_balance_thresh = 50.000000
> nokilluid = 0
> v_pinshm = 0
> strict_maxperm = 0
> j2_nPagesPerWriteBehindCluster = 32
> j2_maxRandomWrite = 0
> j2_nRandomCluster = 0
> j2_minPageReadAhead = 2
> j2_maxPageReadAhead = 8
> j2_nBufferPerPagerDevice = 512
> lgpg_regions = 0
> lgpg_size = 0
> mem_affinity_on = 0
> num_spec_dataseg = 0
> spec_dataseg_int = 512
> lru_poll_interval =
> hd_pendqblked = 380
> psbufwaitcnt = 18030
> fsbufwaitcnt = 13497
> rfsbufwaitcnt = 0
> xpagerbufwaitcnt = 7001
>
> Does anything listed here appear to be causing this.
>
> Thanks.
>
>
>
>

Hi,
You should look at iostat and also filemon output. You may want to
up you numfsbufs (-b) and also your hd_pbufcnt (-B) in vmtune. If
your filesystem is a large file enabled filesystem, you might also want
to look at mounting the filesystem with the "nomind" option to bypass
some of the multi-node buffers that were added as a APAR in AIX 4.3
(where you had to mount the filesystem explicitly with the "mind"
option, but in 5.1 and higher it is the default). If you are not using
large file enabled filesystems you might want to look at using either
the dio or cio option (if you have executables running out of those
filesystems though, you may have some problems), which is like using raw
logical volumes as you are directly accessing the data on the disk. If
your filesystems are on ESS and they are database filesystems, you may
want to look to turn off the queue depth on the disks chdev -l dpo -a
qdepth_enable=no (this will require either reboot or an rmdev -l dpo
-R), then a mkdev on all the devices. Starting with SDD 1.5.x and
higher you are given the option to disable the queue_depth as IBM has
documented in the SDD guide, some database queries are actually hampered
by the buffers.

HTH,

---Phil