Best practices for performance of the Sametime Community Server
Delays can be caused by insufficient throughput of server-to-server connections. Follow these best practices to improve the throughput between servers.
About this task
Occasional delays, especially when data centers from different continents or remote geographies are online and active together, can be caused by large individual messages. If a large message is sent out and the throughput is insufficient, the message can take an unusually long time to be transferred. While one message is being transferred, no other message can be transferred on the same server-to-server connection.
Frequent or sustained delays indicate that the throughput is not high enough between two servers or given servers on remote geographies. In this situation, the delays get progressively longer, until it appears that no messages are being transferred.
Although network bandwidth can be a factor in low throughput, in most production environments, bandwidth is more than sufficient for Sametime®. However, high network latency combined with a small TCP send buffer can often result in network delays. In particular, sites with servers in remote geographies may encounter this problem.
- All Sametime Community Servers and multiplexers (if used) should be located in the same data center. Maintaining a short distance between servers is much more important than the distance between servers and clients.
- If the policy is to deploy servers on remote sites for site redundancy, try to set up sites with very low latency between them. High latency (for example, 250 ms) lowers throughput significantly. Low throughput causes congestion, which in turn causes long delays of 30 seconds or more.
- Here is a simple formula for calculating the throughput of a server-to-server
connection:
throughput = buffersize / roundtrip_latency- Buffer size is 8 KB by default on Windows™. Sametime server-to-server connections default to 64 KB, which is the largest useful size on standard TCP.
- Estimate the roundtrip latency by using the ping command.