ldapadd performance

["Howard Chu" <hyc@highlandsun.com>]

Here are some times running ldapadd against slapd on a PII-400 Linux box,
adding thousands of entries. The machine has 256M RAM and no swapping has
occurred on the system thus far.

back-ldbm
1000 entries
slapd 4.040u 3.030s 4:03.33 2.9%      0+0k 0+0io 844pf+0w
ldadd 0.580u 0.170s 3:22.24 0.3%      0+0k 0+0io 439pf+0w
10000 entries
slapd 50.130u 197.510s 50:16.46 8.2%  0+0k 0+0io 5033pf+0w
ldadd 7.300u 1.350s 46:28.42 0.3%     0+0k 0+0io 1996pf+0w

back-ldbm dbnosync
1000 entries
slapd 12.260u 4.440s 1:13.94 22.5%    0+0k 0+0io 760pf+0w
ldadd 0.700u 0.110s 0:55.86 1.4%      0+0k 0+0io 260pf+0w
10000 entries
slapd 236.250u 102.900s 11:29.12 49.2%        0+0k 0+0io 4866pf+0w
ldadd 6.560u 1.090s 9:05.58 1.4%      0+0k 0+0io 2044pf+0w

back-bdb
1000 entries
slapd 4.050u 13.810s 1:24.73 21.0%    0+0k 0+0io 885pf+0w
ldadd 0.660u 0.160s 1:02.40 1.3%      0+0k 0+0io 260pf+0w
10000 entries
slapd 82.420u 207.640s 21:18.61 22.6% 0+0k 0+0io 2349pf+0w
ldadd 6.270u 1.030s 16:49.83 0.7%     0+0k 0+0io 1998pf+0w

back-bdb dbnosync
1000 entries
slapd 3.530u 0.980s 1:05.26 6.9%      0+0k 0+0io 868pf+0w
ldadd 0.680u 0.050s 0:48.73 1.4%      0+0k 0+0io 260pf+0w
10000 entries
slapd 77.710u 46.730s 17:43.35 11.7%  0+0k 0+0io 2292pf+0w
ldadd 6.660u 1.490s 14:30.48 0.9%     0+0k 0+0io 2131pf+0w

The wall clock times for ldapadd and slapd don't match because I didn't stop
the slapd immediately after ldapadd exited. However, no further operations
were performed. As you can see, the times don't really scale linearly. Also
for back-bdb, the time savings from specifying the "dbnosync" flag is
minimal; since every operation is a logged transaction, there is still a lot
of disk I/O happening on each operation.

  -- Howard Chu
  Chief Architect, Symas Corp.       Director, Highland Sun
  http://www.symas.com               http://highlandsun.com/hyc
  Symas: Premier OpenSource Development and Support