mlock_region failed and kernel parameters

[Steve Post]

I am getting a lot of mlock_region failed messages.
HP9000 V2600 hpux11.00
After restarting sybase, the messages went away. BUT async i/o stopped working for sybase.
Could it be that I don't have a large enough contiguous area in memory for sybase to properly use async i/o?

Should I be changing a kernel parameter? Any suggestions as to which one?
And I already put in the patch that keeps mib2agt from taking all of the memory. That patch works fine.

I looked at webdoc KBRC0001427. Based on that, I THINK I need to lower MAXUSERS from 600 to a lower number (maybe 500?). I want to minimize changes to the kernel because I know how one change can snowball into a mess.

Details
page size=4096
physical mem=12578816 k
lockable mem=10149865 k
avail mem=11644712 k
75% phys mem=09434112 k

aval mem - unlockable_mem = lockable memory
unlockable_mem = avail mem - lockable mem = 11644712 - 10149865 = 1,494,847k
but also .....
unlockable_mem = 200 + (4k/2) * nproc = 9840k <> 1,494,847k

kernel parameters
nproc=4820 = 20 + ( 8 * MAXUSERS )
MAXUSERS=600
swap_on=1 (ie pseudo swap is on)
dbc_max_pct=2
dbc_min_pct=1

sar -v
00:00:00 text-sz ov proc-sz ov inod-sz ov file-sz ov
01:00:01 N/A N/A 171/4820 0 5724/5724 0 1453/9235 0
02:00:01 N/A N/A 218/4820 0 4476/5724 0 1585/9235 0

I also see I am using all the inodes? Perhaps I should increase it? AH. One thing at a time.
ninode 5724


"swapinfo" and "ipcs -mab" is attached.
Steve

[Steve Post]

Well it's been a few days now.

I accidentally posted the message to the wrong spot before. HP removed my original message. Thanks hp. I really don't know how to move my question to the right spot unless I duplicate it. Not the best way to go.

I had one response to use setprivgrp and getprivgrp, on my deleted question. I already have MLOCK privlege for group "users" (aka sybase). That guy should've gotten 5 points for the help (but not the solution).

I called hp and opened a ticket on my memory issue. I was told that a reboot should give me enough CONTIGUOUS memory to get asyc io to work in sybase again. They suggested I try turning OFF pseudo swap.

I haven't really heard anything on this forum post. Did I provide too much info?
Is my question too vague?

QUESTION: swap_on=1. Should I set it to 0?
QUESTION: kernel params listed look ok?
QUESTION: any way to get more lockable memory w/o spending money?

Steve

[RAC_1]

QUESTION: swap_on=1. Should I set it to 0?
QUESTION: kernel params listed look ok?
QUESTION: any way to get more lockable memory w/o spending

swap_mem=1. What it basically does is infalte the value of total swap space available to (swap space configured + 75 % of RAM) This is just for kernel calculations and nothing more. Setting this is good if you have large RAM on the system.

Kernel parameters look OK. The inode setting applies to HFS only. So do you have HFS file systems on your box?? (apart from /stand??)

I do not see any room to get more lockable memory on your system.

Anil

[Steven E. Protter · Accepted solution]

I had one response to use setprivgrp and getprivgrp, on my deleted question. I already have MLOCK privlege for group "users" (aka sybase). That guy should've gotten 5 points for the help (but not the solution).

That was me.

You do need to do that to insure you are getting asynchronous performance.

Other items.

dbc_min_pct=5
dbc_max_pct=7

These figures should be pretty close together because its very expensive in terms of CPU to make changes.

Here is another good document to help with peformance.

http://www1.itrc.hp.com/service/cki/search.do?category=c0&docType=Security&docType=Patch&docType=EngineerNotes&docType=BugReports&docType=Hardware&docType=ReferenceMaterials&docType=ThirdParty&searchString=UPERFKBAN00000726&search.y=8&search.x=28&mode=id&admit=-682735245+1090854105372+28353475&searchCrit=allwords

With regards to the post being deleted, they usually MOVE the posts, they don't normally delete them. I'd check Dan's issues thread(June-July) and see. There may have been a user post indicating your thread should go.

I take issue with anyone saying a post with very good SEP advice in it should be deleted though.

:-)

Good Luck,

SEP