Does DB compact help performance

Question

We have DB where there have been heavy create/delete and read traffic. The DB XML stores around 1.4 million xml documents around 4/8k in size.
The DB XML file has grown to be around 24 gigs.
I understand this is a DB forum but the question seems generic.. given that the DB is a sparse file.. does DB compacting help performance?
We are seeing latency times raise up quite high periodically (potentially correlated with checkpointing).
What kind of information should we collect from the running environment to help deduct the impact of checkpointing on such a DB file?

Answer 1

Using DB->compact can help the performance of any database that has been heavily updated. It will attempt to move data to a smaller number of pages in the database. This will make database page cache more efficient. DB->compact may be run while other activities are happening in the database but it is a heavy weight operation, reading all of the database and updating many pages, so it is recommended that it be used at a time when you have less other activity.
A checkpoint will schedule all dirty pages in the cache to be written to disk. This can certainly slow things down if you have many dirty pages. Using the utility db_stat -m will show you the activity in the page cache. It does not directly show you the impact of a checkpoint but will tell you how many pages have been written to disk. You can use db_stat -mZ to reset the statistics prior to a checkpoint to see more clearly what happens during the checkpoint.
To spread out the effect of the checkpoint you can run a trickle thread (see http://www.oracle.com/technology/documentation/berkeley-db/db/api_c/memp_trickle.html ) which will write pages to disk between checkpoints.
Michael Ubell
Oracle