Description of problem:

Replication is asynchronous, and it is easy to get high replication latency
from dozens to hundreds of entries depending on the topology and system
resources used.
Then, the convergence can take minutes to hours, depending on the case.
Some other products do provide a feature for attributes that must be replicated
as fast as possible.
I do not know the amount of work that such feature would require, but having
some replication latency within more predicable delays could improve
replication and the quality of data provided by replica's (security feature).
May be we already have some other future features about transactions?


Version-Release number of selected component (if applicable):
RHDS8

How reproducible:
RFE

Steps to Reproduce:
1. RFE
2.
3.

Actual results:

examples on a RHEL5.6 x86_64, 2GB ram system with "slow" disk I/O,
redhat-ds-base-8.2.4 in the following topology:

  m1 <> m2  tcp ports 389, 390
    \ /
    hub1    tcp port 391
     |
    hub2    tcp port 392
    / \
  c1   c2   tcp ports 393, 394

example during a 1M entries ADD in round robin to m1, m2:
port #entries
389   33588
390   33576
391   25546
392   25114
393   25040
394   25020
...
389  168113
390  168106
391  168064
392  167997
393  167969
394  167974
...
389  925902
390  925903
391  925885
392  925858
393  925850
394  925849

example during a 1M entries DEL in round robin to m1, m2:
port #entries
389  997428
390  997425
391  997450
392  997475
393  997501
394  997496
...
389  996677
390  996681
391  996716
392  996731
393  996754
394  996755



Expected results:


Additional info: