[915] in Coldmud discussion meeting

root meeting help first previous next next in chain last in chain last

[COLD] More efficient diskbasing

daemon@ATHENA.MIT.EDU (Wed Feb 7 16:31:58 1996 )

Date: Wed, 7 Feb 1996 21:49:41 +0100
From: silovic@srce.hr (Miroslav Silovic)
To: coldstuff@tombstone.sunrem.com
Reply-To: coldstuff@cold.org

It just occured to me that most of the checkpointing time is wasted on
cp - the database has to be copied clean. I thought of a way to keep the
database clean 95% of the time. This would also make the server a lot safer.

The thing that would have to be modified is the cache handling. I suggest
that modified objects could be kept in RAM till the next checkpoint, and
the checkpoint would be when a) 30 minutes (or whatever) passes, or b)
more than 75% of the cached objects are modified. Then we could dump
just the modified objects, and the database would be non-clean only during
the dump, the cp being unnecessary.

What do you all think about this?