Sorry I see there are a couple of posts here awaiting responses.
For reference the documentation is at https://www.ibm.com/support/knowledgecenter/en/SSQL82_9.5.0/com.ibm.bigfix.doc/Platform/Config/c_persistenconn.html
@bcoleman, whether increasing the persistent connection limit slows your relay will depend a lot on your relay’s characteristics. If you have dedicated relays, I’d recommended looking into the Scalability Guide and performing the configurations for a “High-volume relay”, which mostly involves tuning parameters in the TCP stack to allow faster reuse of sockets and expanding the range of ephemeral ports allowed on the relay host. Check the guide linked at https://www.ibm.com/developerworks/community/wikis/home?lang=en#!/wiki/Tivoli%20Endpoint%20Manager/page/Performance%20&%20Capacity%20Planning
Once you’ve configured relay scalability, feel free to experiment with increasing the persistent connection limits and monitor the deployment for results. Please let us know how it works out for you, it’s always good to hear real-world feedback.
@cstoneba, once Persistent Connections are enabled at the relay and the client, the first 3 clients from any subnet to request a Persistent Connection are granted one. Before a persistent connection is established, the Relay sends a UDP notification to the client. If the client receives the UDP notification, it determines that the persistent connection is not necessary and does not establish one. Once a client does establish a persistent connection, it acts as a UDP message forwarder to other clients on its local subnet, so those other clients will receive the UDP messages. A single client with a persistent connection is enough to bring UDP visibility to its entire subnet, we just enable three by default to provide some redundancy; all three forward UDP notifications, so the other clients are less likely to miss a message if one of the forwarding clients is rebooted.
What has been working well for me with several customers is to enable persistent connections on all relays and all clients. Only the subnets that need persistent connections (because UDP is blocked by NAT or firewall rules) establish persistent connections at all, and the “local” clients where UDP is working don’t need or use the persistent connections.
Another note on relay scaling in general, is that the relays are often limited by available network sockets more than processor, RAM, or disk I/O; so an effective technique for some is to put multiple Relays as virtual machines on a single physical host. It’s a way to scale up network sockets without spending on more physical hardware (each virtual relay should have sufficient space for caching being the main cost factor).