samba-2.2.5 performance problem.

[Mick Hellstrom]

People,
I have been using samba-2.0.3 on a K220/2 box with HPUX10.20 for some time now. Performance was quite good with up to 40 developers at one time running Eclipse/ClearCase.

Now I have shifted all this to a R390/2 with HPUX11.00 with samba-2.2.5. The R390/2 is a faster box than the K220/2, however performance has degraded a HUGE amount.

I've fired up glance and found that the disk that handles /var is sitting smack on 100%. Further investigation - it was the samba processes writing continuously to either unexpected.tdb and locking.tdb. NOTE: the size does NOT increase, (as with other complaints about samba). The samaba processes are just whacking these files at a horrenduous rate.

I had to turn of oplocks because performance was REALLY REALLY slow.

Anyone come across this?

Attached is my samba config, (names and addresses changed to protect the innocent).

[Dave Windsor]

I saw a similar problem when upgrading from Samba 2.0.10 to 2.2.4. The problem occurred on both HP-UX 10.20 and 11.x. I was seeing incredible amounts of disk activity around the unexpected.tdb file, and my nmbd process was taking LOTS of CPU time. I sent out a request for help to the Samba mailing list, and got a reply from Jeremy Allison of the Samba Team. He suggested a possible cause, and provided a patch to more efficiently handle the unexpected.tdb. Adding this one-line patch to libsmb/unexpected.c brought my CPU down to a manageable level. Here is the text of Jeremy's email:
========================
Ok, the lseek/read pairs are searching within a tdb for free space on a system that doesn't support mmap correctly.

My guess is that you are being hit with a lot of unexpected udp packets on port 137, you may want to look into what is causing this on your network.

To make the tdb space allocation more efficient you might want to increase the hash size in libsmb/unexpected.c by using the following patch :

Hope this helps,

Jeremy.

=======================

I have attached the patch he sent to me. I have not tried it with 2.2.5 (I'm waiting for 2.2.6). Note that there have been documented problems relating to mmap with Samba on HP-UX.

Good Luck!

Dave Windsor
Robert Bosch Corporation

[Mick Hellstrom]

I have just tried 2.2.7a - same problem. Falls over in a screaming heap when we get mor than around 4-5 developers on it. I'll try and rebuild with the HAVE_MMAP turned off in config.h and see if it makes a diff. Or else I'll just throw my hands in the air and try HP's version, (CIFS/9000).

I can get back to a decent performance if I set the following in smb.conf file:

oplocks = no
level2 oplocks = no
blocking locks = no

This will give me adequate performance.

[Steven E. Protter]

You can NOT run /var any where near 100%

You need extra space on it. The system needs space there to write all kinds of important log data.

If you have online JFS, extend the logical volume and filesystem, immediately.

If not, follow this procedure.

Reboot system, interupt at the 10 second prompt.

type bo

enter

hpux -is

after the system boots, mount the /usr fs

mount /usr

lvextend the logical volume

extendfs the filesystem

cd /
shutdown -ry now

The current version of CIFS for HP-UX is linked below.

You should use this versus the compiled source version. HP's tested it and supports it.

I do not think your problem thouth has anything to do with Samba. Can't run a UX box with var near 100%, the box can stop suddenly and in a catastrophic fashion.

P

[Steven E. Protter]

You can NOT run /var any where near 100%

You need extra space on it. The system needs space there to write all kinds of important log data.

If you have online JFS, extend the logical volume and filesystem, immediately.

If not, follow this procedure.

Reboot system, interupt at the 10 second prompt.

type bo

enter

hpux -is

after the system boots, mount the /usr fs

mount /usr

lvextend the logical volume

extendfs the filesystem

cd /
shutdown -ry now

The current version of CIFS for HP-UX is linked below.

You should use this versus the compiled source version. HP's tested it and supports it.

I do not think your problem thouth has anything to do with Samba. Can't run a UX box with var near 100%, the box can stop suddenly and in a catastrophic fashion.

https://payment.ecommerce.hp.com/cgi-bin/swdepot_parser.cgi/cgi/try.pl?productNumber=B8725AA&date=
P

[Steven E. Protter]

Smack me. Don't give me any points for the Samba part as HP isn't listing depots for CIFS for 10.20.

If the /var part helps, thats fine.

10 lashes for Steve with a wet noodle.

wap
wap
wap
wap
wap
wap
wap
wap
wap
wap

Bad fingers.

Steve

[Mick Hellstrom]

Um. That's 100% disk I/O. Not capacity.