> On Tue, 6 Apr 1999 15:04:36 +0200 (CEST), Andrea Arcangeli
> <andrea@e-mind.com> said:
>
> > I also think that I'll implement the cookie thing suggested by Mark since
> > I am too much courious to see how much it will help (even if my mind is
> > driven by RB-trees ;).
>
> Trees are bad for this sort of thing in general: they can be as fast as
> hashing for lookup, but are much slower for insert and delete
> operations. Insert and delete for the page cache _must_ be fast.
> Expanding a hash table does not cost enough to offset the performance
> problems.
I wrote some sysreq extensions that printk bucket counts for the hash
tables a while back. Want me to dig them up? (I since run my kernels
with hash tables 4x-16x as large -- not sure how much it helps, though.)
I noticed what sct was talking about where when it hits swap, things tend
to all go into buckets close together -- I saw hundreds of things linked
to the same buckets at one point after I made my machine hit swap a bit.
I implemented inode, dentry, and page hash dumping, but I could stick in
buffers too.
Simon-
| Simon Kirby | Systems Administration |
| mailto:sim@netnation.com | NetNation Communications |
| http://www.netnation.com/ | Tech: (604) 684-6892 |
-
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/