RE: Reserved FDs and CPU spikes

> Try increasing your maxusers parameter in your kernel (to
> 512), if this
> is a busy server, you will need at least 8192 FD, I have this
> parameters in
> my kernel:
>
> maxusers 512
> options NMBCLUSTERS=32768 #Support for 32K for MBufs
>
> Normally my server uses ~3,000 FD, peak reported ~12,000 FD, I have
> available 32,768 just in case.

 What would you define as busy server? At the time of these problems,
 I see 1500 requests per minute.

 Also, I ran the same hardware with Redhat 6.2 and Squid 2.3STABLEsomething
 under the same user environment successfuly for a couple weeks without
 seeing FD problems. Redhat defaults to only 1024 FDs available, too!

 Has something significant in the allocation of FDs changed between 2.2 and
2.3?

 Is it normal for Squid to suck up a ton of FDs as "reserves" ? Prior to
 this problem, I'd not been aware of "reserve" counts much over 100.
Received on Fri Jun 02 2000 - 07:57:09 MDT