On 4/14/05, Daniel Salama <dsalama at user.net> wrote:

> I apologize for the mis-reading of the BTree random read result. It was
> not 10 seconds. It was 1.02 seconds. However, I noticed something
> peculiar. I ran 4 times a piece of code which simply read the 6666th
> element of the BTree tree and in all 4 times, it gave me different
> execution times (2.9 secs, 0.9 secs, 1.3 secs, and 0.8 secs). During
> all these, I let Squeak sit idle for a while and then ran it again.

I can't speak for Magma, but with GOODS if you always use the same
connection and don't #flush you'll find that the 2..nth read is much
quicker than the first one - the time is spent in bringing in the
needed objects, not in doing up the lookup once they're there. 
Looking for other objects that are nearby should also be quicker (and
the effect will be less apparent the further away you look).

Avi