Re: [PATCH] Split futex global spinlock futex_lock

[Jamie Lokier]

Ulrich Drepper wrote:
> After these changes the code still works but I couldn't really measure
> any differences to the code without the extra attributes.  This is on a
> 4p machine with 10 processes running in concurrently using mutexes and
> condvars with 250 threads each.  This might be because either the hash
> function is good or very bad (i.e., hashes all futexes in the same
> bucket or far away).  I guess the extra attributes don't hurt.

On a 4 CPU machine the hash table is plenty large enough without the
extra attributes.  (Assuming 128 byte cache lines: 3072 bytes without
(24 cache lines); 32768 bytes with).

The extra cache lines might hurt a bit when all the threads run on a
single CPU, or on a HT, just because 32768 bytes is a lot more L1 than
3072 bytes.

-- Jamie
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/