Smarts SAM notification messaging latency and maximum throughput
When Smarts SAM receives a notification from an underlying server, it sends a message back to that server to get additional information. When that additional information is returned, SAM may find that it needs more. This can happen as many as 3 times per notification one way, and assuming that the network latency is bidirectional, this means that the network latency will be encountered 6 times. For example, if there is a network latency of 39 ms, Smarts SAM has a minimum .234 seconds (39 ms * 6) delay caused by latency alone to process just one notification. Even if no other factors are taken into consideration, that limits SAM to a maximum of just over 4 notifications/second for this particular data path.
Factors limiting actual SAM notification throughput
Actual throughput on a system with a network latency of 39 ms will probably be much lower than 4 notifications/second. This is because a SAM server often has to do other tasks such as calculating impacts, getting notifications through other data paths, executing scripts, scheduling escalations, topology synchronizations, and database saves. And, a SAM server may not be connected directly to the underlying server. If there are other servers such as aggregating SAM servers in between, then there will be even more network traffic and latency to contend with.
Impact of server queueing on notification delays
A notification throughput rate of less than 4/second will probably acceptable and may not even be noticeable as long as the notifications are coming IN at a rate that is lower than this throughput rate. However, as soon as the rate of incoming notifications exceeds the throughput rate, SAM and the underlying server or servers start placing items in queues. If the notifications are coming in faster than 4/second with this scenario, the queue will grow. The longer the queue grows, the longer it will take for a notification to reach SAM. Consider that if the SAM is only handling 4 notifications/second, a queue with only 10,000 items in it (which is not usually considered particularly large) will take 2500 seconds or just over 41 minutes to be completely processed.