Push vs. Pull Transaction Replication

Q: Push vs. Pull Transactional Replication
We are currently utilizing a push transactional
replication of a subset of the tables in a production
database involving 3 machines; the publisher which is our
production db machine with 4 hyperthreaded processors, 8
GB ram, a distribution db machine with 1 processor, 2 GB
ram, and a subscriber db machine with 2 hyperthreaded
processors, 6 GB ram to which this table subset is
replicated. All machines are SQL EE with SP3. The
subscriber is used for complex searches and has 6 indexed
views resident on it. All replication agents are set to
run continuously.
The Distribution Agent profiles have been left at the
defaults except for QueryTimeout, which has been set to
3600.
Because we cannot simulate this environment for testing,
I would appreciate hearing from anyone who has experience
with a push vs. pull subscription from a performance
standpoint. Thanks very much for your input.
pull offloads the distribution agent to the subscriber. Hence your
performance on your publisher is better.
The problem is that then you no longer have a central point of
administration - in other words you manage your publications and snapshot
agent on your publisher, but your distribution agent on your subscriber.
I can't quantify exactly what performance increase you will have on your
publisher as it is largely a function of load and hardware on your publisher
and subscriber.
You will have to test to see what performance increase you get. You can get
an idea of what it will be by stopping your log reader and distribution
agent on your publisher.
"fundster" <anonymous@.discussions.microsoft.com> wrote in message
news:98dc01c433b2$498477f0$a301280a@.phx.gbl...
> Q: Push vs. Pull Transactional Replication
> We are currently utilizing a push transactional
> replication of a subset of the tables in a production
> database involving 3 machines; the publisher which is our
> production db machine with 4 hyperthreaded processors, 8
> GB ram, a distribution db machine with 1 processor, 2 GB
> ram, and a subscriber db machine with 2 hyperthreaded
> processors, 6 GB ram to which this table subset is
> replicated. All machines are SQL EE with SP3. The
> subscriber is used for complex searches and has 6 indexed
> views resident on it. All replication agents are set to
> run continuously.
> The Distribution Agent profiles have been left at the
> defaults except for QueryTimeout, which has been set to
> 3600.
> Because we cannot simulate this environment for testing,
> I would appreciate hearing from anyone who has experience
> with a push vs. pull subscription from a performance
> standpoint. Thanks very much for your input.
>
|||I would think that you would not need to stop the log reader. Even with a
pull subscription the log reader is still running on the publisher.
Rand
This posting is provided "as is" with no warranties and confers no rights.
|||that's a very good point Rand, I missed that.
I was trying to get a baseline "without" replication, but to get a
representative baseline with pull subscriptions you really should have the
log reader agent running.
"Rand Boyd [MSFT]" <rboyd@.onlinemicrosoft.com> wrote in message
news:nxrqWlENEHA.3780@.cpmsftngxa10.phx.gbl...
> I would think that you would not need to stop the log reader. Even with a
> pull subscription the log reader is still running on the publisher.
> Rand
> This posting is provided "as is" with no warranties and confers no rights.
>
sql