Monolithic system, max_dbc_pct (during Samba upgrade)

[Ryan Novosielski]

Our system is a 6-way N-class with 2.50GB of physical RAM. The system is running Apache, Oracle (though Oracle is barely seeing any use), print spooling for about 15 printers, mail for a few thousand users, NFS for about 10 workstations that see light use, and is running Samba for 2 or 3 computer labs with about 30 computers each, along with the usual shell/FTP/POP/etc. protocols, along with some light SAS use.

I tried to upgrade to Samba 3.0.9 from 2.2.12 and Samba says in its log fairly consistently: "open_sockets_smbd: accept: No buffer space available." My assumption is that this is a result of insufficient buffer cache available.

Right now, dbc_max_pct is set to 16, or roughly 409.6MB. I feel that this value is low, as many of our filesystems run at or near 100% utilization, and we are attempting to host a relatively busy new application (newer version of Samba). What SHOULD a value for a server like this be? The system often runs north of 75% for memory usage (and I'm told was hovering at about 97% yesterday). I do not want to cause the machine to swap like mad. A memory upgrade is probably in order, but I want to know the proper value for that setting in order to know how much RAM to buy.

Any insights? Thanks!

[Ryan Novosielski]

Forgot to mention this machine is also an Omniback Cell Manager and device server... however, the backups take place at night (not when I am having this problem).

[Bernhard Mueller]

Hi,

double your RAM to 5GB, then 16% dbc_max_pct would amount to 800MB and you would get 2gig memory on top as a safety margin.

my 0.02$

Regards,
Bernhard

[Victor BERRIDGE]

Hi,
Truly with what you just said the box would need at least 4 GB of RAM to be able to do the task correctly

More would be better of course...

All the best
Victor

[Victor BERRIDGE]

Having said that,
I dont believe it has something to do with dbc_max_pct...
Do you have glance installed? Are you using X?
let it run for a while... gpm&
And look at the alarm history, and check at the system tables report to see if you havent reached some upperlimits...


All the best
Victor

[Geoff Wild]

I think your buffer cache is fine. What do you mean 100%? cache hit ratio? then that is great!

If you are hitting 97% memory usage - don't increase your buffer cache...

Check out this thread:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?admit=716493758+1079078265934+28353475&threadId=493734

It does sound like your server may be overloaded.

I run Samba 3 on a few systems (they have a lot of ram) as well as my little workstaion (B2000 with only 768MB) - which also runs MC/SG in test.

Rgds...Geoff

[Ryan Novosielski]

No no, the use of the disk... and the disk bottleneck probability is at WRN and often CRT throughout the day. Not 100% cache hits.