[Date Prev][Date Next] [Chronological] [Thread] [Top]

Slow delete behaviour



I'm playing with Ldap-2-devel (latest as of a week or so ago) with
Sleepycat DB (last v2) and NT4. Adds to LDAP work fine (~150/sec on
466 MHz Celery, IDE disk) with just a few indexes and searches also are
fast. But I find with deletes there is a *warm-up* period of disk
thrashing before the delete happens that increases with the number of
entries in the database.

For example, with 10K test entries the disk thrashes madly for 10 secs
before the first delete happens,  then the subsequent ones go thru
merrily enough at about 2/sec. With 100K entries, the disk thrashes
madly for 10 MINUTES before the first delete happens, the others then go

thru about 1/sec, which is ok. If I kill the client at this point and
restart, exactly the same thing happens. There is no VM activity or much

CPU usage at all, just disk I/O, seems to be scanning something.

What on earth could be causing this *warm-up* thrashing? Why would it
happen on only the first delete in a session? Sounds to me like it is
scanning some index first, but why?