3 ms·
I've taught queuing theory before, and my favorite part is teaching about cumulative flow diagrams. I will preach it to anyone who will listen. The CMF is the
by jedberg 3y ago
I've taught queuing theory before, and my favorite part is teaching about cumulative flow diagrams. I will preach it to anyone who will listen.
The CMF is the single best way to monitor your queues in your monitoring system, and most people don't do it. Almost everyone has a graph showing the current queue depth. They might even have alarms on it.
But if you see the queue depth going up, how do you know the cause? There are two possibilities: arrival rate increased quickly, or service rate dropped quickly.
But with a CMF, you can tell just by looking! One of the two sections will be getting fatter, telling you immediately where to look for the problem. If service rate is slowing, then you know to scale up your processing fleet, or look for a pathological case that is blocking processing. If the arrival rate is increasing, you can put in some backoff or load shedding, or just be happy about how busy your product is!
[0] https://en.wikipedia.org/wiki/Cumulative_flow_diagram https://en.wikipedia.org/wiki/Cumulative_flow_diagram
- dpc_01234 3y agoI wanna know, please send links senpai. Edit: Couple of YT videos told me everything I need to know, I guess? A bit sad it seems mostly used for "Agile".
- kqr 3y agoReading out the delay as a leading indicator is also a neat thing you can do with cumulative flow diagrams. If you have stacked CFDs, you can immediate spot when a stage is the bottleneck, or when flow is jerky across one of the interfaces.
- dang 3y agoIs http://brodzinski.com/2013/07/cumulative-flow-diagram.html http://brodzinski.com/2013/07/cumulative-flow-diagram.html good? If so, I'll send a repost invite for https://news.ycombinator.com/item?id=18903965 https://news.ycombinator.com/item?id=18903965. Then please come back and post about it again :)
- jedberg 3y agoCFDs are almost always used for kanban and development, so almost every article is about that. That article is perfectly good in that regard. I've never actually seen an article about using them for operational monitoring of queues. I've only ever given talks about them. Maybe I should just write one... But if you do invite that article to repost, I'll comment on it anyway!
- dang 3y agoHow about you write one? That sounds much better! If you do, and are willing to email hn@ycombinator.com, I'll be happy to give it the SCP treatment (https://news.ycombinator.com/item?id=26998308 https://news.ycombinator.com/item?id=26998308).
- jedberg 3y agoWell how could I turn that down? I'll let you know!
- javier2 3y agoAny way to do that with grafana?
- rbjorklin 3y agoProbably just a stacked line graph?
- smadge 3y agoAre cumulative flow diagrams useful for serving systems, where arrival rate > departure rate causes immediate user visible issues?