[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
back-hdb
The new Symas-developed hierarchical database is now available for testing in
CVS HEAD. The configure script has not been updated yet so you'll have to
enable it by hand. Full support of subtree rename is not yet implemented in
the cache manager, so don't try it yet.
My initial tests show that back-hdb is a only few percent faster (~5%) than
back-bdb for a read-only set of operations. This is to be expected;
back-hdb's database usage is more efficient so it starts up faster but they
both use the same cache code. Once the entries are cached the difference
between them averages out to zero over time.
However in a read/write mix, back-hdb is at least 20% faster than back-bdb,
and generally runs with 50-75% fewer lock conflicts and deadlocks. This is a
big win for concurrency. This performance is fully 6 times faster than
back-ldbm in the same test. (For a read-only test, back-hdb is only ~2%
faster than back-ldbm.)
If you have some benchmarks of your own to try, please post the details and
results.
-- Howard Chu
Chief Architect, Symas Corp. Director, Highland Sun
http://www.symas.com http://highlandsun.com/hyc
Symas: Premier OpenSource Development and Support