Class N performance

[Abdel Guelmamene]

Hi,

I've a problem with my N400 (4 processor, 4 GB RAM , FC acces to XP512 disk)

I have only an Oracle database 8.1.7 in my system. But all time the glance give me

GlancePlus Current Avg High
CPU Util 20% 21% 21%
Disk Util 100% 100% 100%
Mem Util 65% 65% 65%
Swap Util 57% 57% 57%
----------------------------


My kernel parameters :

dbc_max_pct 15
dbc_min_pct 5
max_thread_proc 1024
maxdsiz 0X10000000
maxdsiz_64bit 0x0000000040000000
maxssiz 0X01600000
maxssiz_64bit 0X01600000
maxuprc 256
nfile 2292
nflocks 300
maxfiles 1536
maxfiles_lim 2048
shmem 1
shmmax 0X60000000
shmmni 1024
shmseg 32
semmns 1024
semmni 512

The sar -b 1 12
15:26:21 %rcache %wcache
15:26:22 77 78
15:26:23 67 44
15:26:24 70 70
15:26:25 63 97
15:26:26 80 99
15:26:27 73 65
15:26:28 70 83
15:26:29 78 70
15:26:30 78 54
15:26:31 64 100
15:26:32 71 81
15:26:33 77 29

Average 72 81


and the resultat off sar -u 1 12

15:27:45 %usr %sys %wio %idle
15:27:46 18 14 67 1
15:27:47 17 15 68 0
15:27:48 17 15 68 0
15:27:49 15 23 62 0
15:27:50 16 15 69 0
15:27:51 19 24 54 3
15:27:52 17 20 63 0
15:27:53 19 18 63 0
15:27:54 16 20 64 0
15:27:55 16 16 68 0
15:27:56 18 19 62 1
15:27:57 18 20 62 0

Average 17 18 64 0

I dont understand wat's the problem with my system!!!! is it my Kerne? My hard? My Oracle?

Please i need help, and excuse me for my broken english.

Regards

[Hai Nguyen_1]

Can you post the output file of the two commands below:

# kmtune -l -q timeslice > /tmp/output.txt
# top -d 1 -n 1000 -f /tmp/output.txt

Hai

[Abdel Guelmamene]

Hi Hai,
This the informations you have requested:

kmtune -l -q timeslice > /tmp/output.txt

Parameter: timeslice
Parameter: timeslice
Value: (100/10)
Default: (100/10)
Minimum: -
Module: -
----------------------------

Regards

[Hai Nguyen_1]

Abdel,

I have reviewed the output you attached. Your CPU timeslice is correct. In the top output, the memory utilization per process seems to be normal as well. Based on your glance output, your disk(s) appears to be always busy and this may have something to do with the busy I/O of the Oracle instances on the server. From my experience with glance, you can trust the reports on CPU, memory and swap utilizations. As for disk utilization, NEVER trust glance since you will ALWAYS get a report of 100% disk utilization even though only ONE disk of MANY on the server is really busy.

Anyway, my general comment is your box is working normally.

Hai

[Tim D Fulford]

What does "sar -d 12 1" show?

The 100% disk utilisation could be from 1 or more disks. A knowledge of what disk contains what would be useful too. e.g.
vg00 - HP-UX, c1t6d0 c2t6d0...
vglog - redo logs etc c3t1d0, ...
vgdat - data disks .....

You can also look at the disks historically using MeasureWare. (i'll try to root out the a previous thread, it is somewhere in my answered q's section)

Regards

Tim

[Tim D Fulford]

Here is a thread I posted earlier

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x7dc13a7b3682d611abdb0090277a778c,00.html

service time =~ disk%*10/Phys IO rate

Tim

[Rajeev Shukla]

Can you see which disk is actually hitting 100% and then focus what is on that disk. Try then moving things from that disk to another because sometimes people configure everthing on a single disk which is bad for Databases. In sence the logical logs should be on a separate disk, and the most accessed database on one disk and hence forth. Try to distribute things of that disk which is 100% to other disks.

Cheers
Rajeev

[Tim D Fulford]

I suspect you have a couple of things that may be going on. I use Informix so some things in Informix may translate directly across, others may not.

1 - Lots of sequential scans, maybe your statistics are not up to date. SQL will be required to see if this is so.
2 - Not enough IO threds. Informix uses two types KAIO (Kernel Asynchronous IO) which actualy means it's processes hook into the Kernel & let the kernel do the IO. The second is aio-vps, which are simply processes that perform the IO and you need lots say 30+. I suspect that Oracle uses one or both of these types of methods, probably named differently.
3 - As mentioned above the layout of data on you disk is not good. Of the articles I've read Oracle like to use SAME (Stripe And Mirror Everything). This is really quite effective method of getting the most out of your disks with out a great deal of trial and error. IF the disks look like they have "HOT SPOTS" (single busy disks limiting the performance), This method is really quite a good way of ensuring you get the most.
4 - The disk subsystem is not configured correctly. If you are using RAID 5 you could become stuck if you are performing lots of small random IO's.

I hope the above gives you some pointers.

Regards

Tim

[Stanimir]

Hi!
Please show me the value of your Kernel`s
paramater:

unlockable_mem = ?

Regards

[Abdel Guelmamene]

Hi Stanimir,

The value of the paramete you ask is 0
kmtune |grep -i unlockable_mem
unlockable_mem 0


Regards