import problems (mostly) fixed

So M10 has been working towards killing the last "showstopper" memory
bug... namely the inability to import a massive tree (such as when
BenC tries to import the linux kernel... he runs out of swap!)

Here's my original test case:

  Importing emacs source tree: 86 megs, 3000 files & dirs.

  * import: 290 megs of RAM used. (sending inline txdeltas)

  * add/commit: 108 megs of RAM used, then stayed there as postfix
                 txdeltas were sent.

It -seems- that the problem was that there were simply too many pools
being created and held around. Remember that each pool, when created,
is a *minimum* of 8K. Gstein's latest commit (r1456) has drastically
reduced the number of these pools.

   As a result, my import now uses only 16 megs.

In general, something is still vaguely wrong; even 16 megs is too
high. But this is generally passable for M10, which we'll start
rolling tomorrow.

Ben Collins: you might want to try re-doing your linux kernel import
and see what you get.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Mar 7 23:51:31 2002