We run a large number of publications and subscribers for load balancing of heavily accessed data across servers. We use a separate distributor with processing power to do this. However we have recentlycome across a problem where there seems to be a limit to the number of distribution/snapshot agents that can be running. This initially looked like a worker threads problem so we increased this but it had no impact. Agents go through step one ('starting agent' message) but then appear to just hang. I've profiled the distributor and there is no activity there at all while the agents are hung. There are no locks anywhere.
I'm wondering if there's a limit on the number of remote server connections allowed at one time but have found no info on this.
We currently have 112 distributor processes running and 4 logreader processes along with the periodic housekeeping agents. Any further agent added just hangs unless I stop one of the others.
Would be interested in any ideas anyone has on this.