7500 stat entries sounds plausible. I'm surprised by the "500 dcache
entries", though.
At 05:17 PM 1/13/98 -0500, Chuck Lever wrote:
>
>lws+ says:
>< Chuck Lever <[EMAIL PROTECTED]> writes:
>< > that's very interesting. i found that increasing the size of the
>< > dcache hash table larger than 500 caused degenerative hash behavior,
>< > resulting in degraded performance (because the kernel spent more
>< > and more time adding and searching entries in a few hash buckets
>< > instead of spreading the entries over the entire table). this is
>< > also apparently true of the stat hash table.
><
>< Of course, it's a case of diminishing returns. We made these hash
>< tables larger in 3.4, and LRU the hash buckets, so if you measured 3.3,
>< your observations may be somewhat outdated. In the limit, you're still
>< correct, but the limit has moved.
>
> i measured 3.4 and 3.4a about a year and a half ago. the limit was
> about 7,500 stat entries.
>
>--
> Chuck Lever - [EMAIL PROTECTED]
> U-M ITD Login service team
>
