Deadlocks in merge replicated database server.

We are using filtered merge replication between server and clients. Previously all the clients used to update local database and then replication used to relay those to server to other clients. Recently to reduce traffic, since 99% of the time clients are connected, we decided to change the architecture so that the clients update the server directly and changes are downloaded and local db is kept up to date in case it goes offline. The problem is that since we changed this, we are getting lots of deadlocks.
The filter table is the bottleneck and is involved in most the deadlocks. This table is queried by all the merge objects and is directly modified by the code as well.

---

Regards Mayur, MCTS Windows,Web and Distributed applications