Re: nfsd problems with FreeBSD 5.2.1

Eric Anderson wrote:

I'll go through one by one on this.

> Hi Mike!  I've had some experience with FreeBSD/NFS/NIS and might be 
> able to help.. (maybe!)
>
> Have you done a tcpdump to see what traffic is like during that time?


Yeah, the packets come in very quickly from several hosts, all via tcp.

>
> Also - is your NIS server that this mail server is connected to being 
> used by a lot of clients?  I've seen very strange hangs, delays, and 
> loads when the NIS server doesn't respond quickly enough..
>

The NIS server doesn't have a lot of clients, less than 10 off the top 
of my head, nothing that would cause it to be very slow with responses.

> Just for your info, what I use on my clients (I have about 800 NFS 
> clients), is UDP, NFSv3, 4196 rd/wr size, hard mount, interruptible..
>
> Does mountd have a high run time? Maybe it's a mount storm (if you use 
> automounter on your clients)
>
Mountd has consumed 0:01 minutes of cpu time with the machine being up 
two days, since I know of every machine thats mounting the nfs share, I 
have visited each one individually to make sure the mount was correctly 
accessable and such (since I was playing with mount options)

> What about network interface errors?

To be honest, i'm not sure how to check for these, in linux/solaris, 
ifconfig itself prints out the errors, this isn't the case (at least 
with ifconfig -a) on FreeBSD, as far as I can tell.

>
> If you watch an iostat during the busy time, is your disk being slammed?

iostat seems to never go over 30k/sec average,

>
> If you are using uw-imap, how big are the mailboxes?
>
See, this is the thing that I think may be causing the problems. We are 
using uw-imap, and the mailboxes vary anywhere from 100k to 600mb (some 
people are irresponsible mail users!) The thing is though, even in times 
of high load like someone opening or accessing a mailbox of this size, 
it shouldn't be causing the box to go down to its knees, should it?

I've recompiled the machine with 5.2.1-release-p9 and disabled 
hyperthreading, we'll see how that goes.

--Mike