Shared or Independant Distribution Agents?

  • We have Lots of databases wich typically have three Trasactional Publications. Some DB's use a shared distribution Agent for all three Pubs whilst some DB's use Independent Agents. The reason for this set up seems to be purely by chance.

    BOL says "Sharing the same Distribution Agent reduces the resource demands and increases performance at the Distributor". Whilst some resources on the net say "IF you use independant Agents, you can get performance improvements while replicating your data". This seems to be conflicting information.

    I know I should go with BOL but having seperate distrubition streams does make sense (althoguh I now have three jobs instead of one using this method). What do other people think?

    Thanks in advance

  • Network configuration would probably impact this heavily. If you have 10 distribution agents and enough bandwidth between your publisher, the distributor, and the subscribers you may see better performance in the replication overall - although this would increase the load on the publisher.

    I have rarely seen the distributor as the bottleneck in a replication performance problem. It is almost always bandwidth or simply transactions on the publisher that are not optimized to be replication-friendly. So, I would tend to want to consolidate as much as possible just to make management easier.

  • hi ,

    so you are saying go with the Shared distribution Agent?

  • Unless you are actually seeing the distributor getting resource-bound, that is what I would recommend.

  • It also depends on how many subscribers you have perpublication. If it is 1-to-1 shared is probably Ok but if it is 1-to-many indepenedent can give you "flexibility" in management.


    * Noel

  • We never have more than two subscribers.

    But we do have at least three Trans Publications per DB. And yes it can get confusing sometimes as you see the same Distribution Agent messages under any of the Trans Publication views!

    As I am migrating about 20 Replicated DBs per server I have decided to go with this approach....If it already has shared Dist Agents then keep it this way. If it has Independant Agents then Keep it this way.

    When in doubt go with what is proven to work already! 😎

    Thanks for you help...

  • An important point to note, if one publications fails with dodgy data then this affects the other Publications if using Shared Distributor! Is this alone a valid case for independent Agents?

  • schwizzla (6/23/2008)


    An important point to note, if one publications fails with dodgy data then this affects the other Publications if using Shared Distributor! Is this alone a valid case for independent Agents?

    That is one of the points I tried to make when I mentioned "flexibility". It will let you operate on a "subset" of the entire metadata infrastructure. So you will have to worry about fixing/updating/rescheduling ONLY the agents/publications you are interested in!


    * Noel

Viewing 8 posts - 1 through 7 (of 7 total)

You must be logged in to reply to this topic. Login to reply