cycbuff limit?

[Pete Ashdown]

* Daniele Orlandi (daniele at orlandi.com) [000825 16:28] writeth:

>> Ok, the linux kernel I'm using now supports 64K fd's.  I set ulimit and
>> rlimitnofile to 32768.  It still has the same behavior.
>
>Pete,
>
>There are two limits, one is the total number of open fs the kernel
>maintains, the other is a per-process limit. Older linux kernels had 256
>fd per process limit, while newer ones have 1024. The limit is hardcoded
>in one header file, so you might need to recompile inn if you changed
>the kernel.

I recompiled.  Still has the same behavior.

>As a rule of thumb, the per-process limit should be the number of
>cycbuffs plus an overhead of 30-40 fds (someone will be more precise on
>this), while the system limit should be the per-process limit multiplied
>by the etimated peak of the number of clients.

That would be far lower than 1024.  I'm running kernel 2.2.16.  What's up?