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

back-hdb and HEAD



After much work by Howard (coding, and reading profiling data) and me (benchmarking HDB/BDB and gathering profiling data for Howard to evaluate), HDB is now equivalent in performance to BDB.

This is a *substantial* gain over where HDB was previously. In the initial tests, in a pure 100% read scenario, HDB averaged 5.996 operations/second when having a subtree as a base. BDB averaged 7,050 operations/second.

Current Read & Read/Write tests show:

100% READ -

BDB: 7,060 ops/second
HDB: 7,127 ops/second

80% READ, 20% WRITE -

BDB: 4,567 ops/second. 912 ops/second MOD, 3654 ops/second SEARCH
HDB: 4,491 ops/second. 901 ops/second MOD, 3591 ops/second SEARCH

50% READ, 50% WRITE -

BDB: 2,083 ops/second.  1,041 ops/second MOD, 1,042 ops/second SEARCH
HDB: 2,111 ops/second.  1,055 ops/second MOD, 1056 ops/second SEARCH

20% READ, 80% WRITE -

BDB: 1,336 ops/second.  1,069 ops/second MOD, 268 ops/second SEARCH
HDB: 1,363 ops/second.  1,091 ops/second MOD, 272 ops/second SEARCH


At this point, I would highly recommend the latest avl/tavl changes in the liblutil area, and the bits in servers/slapd/back-bdb get synched from HEAD to 2.3 so that it can enjoy this substantial improvement. ;)


--Quanah

--
Quanah Gibson-Mount
Product Engineer
Symas Corporation
Packaged, certified, and supported LDAP solutions powered by OpenLDAP:
<http://www.symas.com>