perryibm commented on issue #2569: 1 million per second event processing using openwhisk ?? URL: https://github.com/apache/incubator-openwhisk/issues/2569#issuecomment-320257008 1. It matters if you are considering steady state rather than a burst because it is then not a latency requirement but a bandwidth one. Assuming you want to continuously process 1 million events/second in steady state, then the amount of time the event handler takes times this rate is the computational load the system needs to handle because in a steady state assumption, over any longish window, the drain rate must equal the injection rate. But, if the load is spikey then another way to phrase the question would be at what larger time window can you determine an upper bound on injection rate. If you don't care about latency at all over any time window, then see below. The limit on the number of concurrent invocations is determined by a system setting so it's up to a sysadmin to determine. The queue lives in an internal Kafka. If the invocations are asynchronous and you don't mind a really high latency before service, then one could turn this up. For example, you could set up a system that could actually only handle 100K events/second in terms of actual computation, then a burst of 10 seconds at 1M/second will cause the queue to increase to 10 * 1M - 10 * 100K = 9M which will then take 90 seconds to drain. The gap between 100K and 1M is large and this example is a little extreme but if you actually know that is this burst happens say no more than hourly, then it's fine to choose such a config. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
With regards, Apache Git Services