I've just updated my LMS to use the latest DBD::sqlite (1.58) and could
run some benchmarks, then downgrade to 1.34 and re-run them.
The question is how to benchmark (re-scanning the library is simple, but
not really representative of normal use) - how did you measure the 20%
performance gain?
I only tested the scan. It's very heavy on database stuff, with a good
mix of read and write. When scanning the collection is 20% faster then
something has been improved. Either optimized query paths, processing of
data, or I/O or whatever.
Running this in a loop
Code:
--------------------
while true; do ./bench-lms.sh; done
--------------------
puts 100% load on the server and should be fully DB-bound.
Agreed, benchmarking is difficult to do correctly. Running this kind of
stuff in a loop would likely only hit SQLite's buffer - which is not
representative either. In general I've had little reason for complaints
with my setup in day-to-day use. But I know that eg. some of Erland's
plugins are extremely heavy on database stuff. If you're running one of
those, that might be interesting, too.
One other reason why I was looking into updating SQLite is that it
supports new features in the fulltext indexing I'd like to leverage.
--
Michael
_______________________________________________
discuss mailing list
[email protected]
http://lists.slimdevices.com/mailman/listinfo/discuss
