On Thu, 10 Feb 2000, Chuck Lever wrote:
>hardware. but on large memory hardware, large caches pollute the CPU
>caches during searches, actually slowing the system down.
Yes. As I was saying to Linus privately there are two problems:
1) avoid CPU hanging at each `ls` due too much hash collisions
2) load balanced releasing of dynamic objects due memory pressure
(1) and (2) are completly orthogonal.
I did (1) that is currently necessary unless we want to allocate hashtable
large enough to have a mean distrubution of 1 if 200 giga of RAM are
allocated in dcache _or_ icache (that would harm memory usage for a quite
uncommon case). The fixed percentage has to match the dynamic hashtable
algorithm that chooses the size of the hash (and it make perfect sense
since the size of the hash if fixed too).
I did design for (2) too (that should be done anyway on the long term) but
it's not a one liner and first I have to finish some other thing before
starting implementing the thing.
(1) has the (for now good) side effect of hiding well the lack of (2).
Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Feb 15 2000 - 21:00:17 EST