[Samba] Performance Issues With Samba and RedHat 8.

[Ross McInnes (Systems)]

As an aditional note/something ive just noticed, during these "slow downs" 
(one of which is happening as i type) we quite often get a "user"(random 
not something they are doing thats naughty) who sits on a high cpu smbd 
thread.

like so :

19566 nm102     15   0  4116 3860  2592 S    15.1  0.1   0:02 smbd

the 15.1 % is the cpu usage, and its been sitting there now for about 10 
mins.

any ideas why or what it could be doing?

also at the same time the samba log and an increased amount of OPLOCK 
errors/requests. related?

Thanks

Ross

On Tue, 6 May 2003, Ross McInnes (Systems) wrote:

> Appologies if this has been asked before, i have searched the archives and 
> while there are some almost related issues, none really answer my 
> question/problem.
> 
> Recently (about 4 months ago) we purchased a new Dell 2600 (2.0Ghz Xeon 
> 2GB Ram), Machine (single cpu) and migrated the data and system from rh 
> 6.2 to 8 on this new server. By migrate i mean Installed rh8 fresh and 
> moved the user data and .conf's, cron etc.
> 
> All was well and fine until one afternoon we had a massive slow down in 
> the "system" (overall system, network etc. not this actual server). upon 
> investigation we noticed that smbd was pulling 99% cpu time for about 
> 15-20 mins (before we decided to restart smbd)
> 
> restarting smbd, we had a small issue where all the old links etc tryed to 
> contact the server so a little slow for a few mins then everything went 
> back to normal. Very odd and certainly something we had not previously 
> experienced on our older server.
> 
> so we kept an eye on it for a few weeks and all seemed well. we had a few 
> reports of the system being incredibly slow at times, although this did 
> not coincide with smbd using 99% cpu. using IPtraf/ethereal/ping showed 
> that there wasnt a network issue to or from the server. 
> 
> Looking at the samba log file we started to notice a lot more OPLOCK 
> requests/bad/failures etc around about the same time everything was going 
> slow. So on a few of the shares we made them read only and put on FAKE 
> OPLOCKS (netlogon, start menus etc). And we changed level2 oplocks to no. 
> 
>  # Turn off level 2 oplocks
>         level2 oplocks = no
> 
> This seemed to improve things until it went slow/99% usage once again. 
> (about 2 week or so later) Hair ripping time once again.
> 
> So after more research I read an article about the smbd.conf and some 
> sort of windowing problem with samba and windows clients. so i modified 
> "socket options" to be....
> 
> # Most people will find that this option gives better performance.
>   # See speed.txt and the manual pages for details
>   # You may want to add the following on a Linux system:
>   #         SO_RCVBUF=8192 SO_SNDBUF=8192
> socket options = TCP_NODELAY IPTOS_LOWDELAY SO_SNDBUF=8192 SO_RCVBUF=8192
> 
> Now this seemed to work and all was well until the past 3 weeks, where we 
> have had to restart the server twice (no console response) and have had 
> numerous problems with 99% cpu on smbd or this incredible slowdown of 
> loging on, opening files, listing dir's etc.
> 
> Other details about samba is that its version 2.2.7a (according to 
> smbstatus), we have nt4 clients and on average have about 100-150 users 
> at peak times. it also has the wins server option enabled (so that it acts 
> like a wins server)
> 
> I have no other ideas to try or even think to search for, so any advise, 
> questions and maybe answers would be gratefully received.
> 
> Faithfully
> 
> Ross McInnes
> 
>