If you have somebody doing work that can appear at random (like somebody calling and saying they have a problem), that person will either be free for a fraction of time that seem high to naive people, or will have a line and take ages to help anybody approaching them.
That seemingly high fraction of time is usually around 50% for the line to stay under control. That’s a well known result from mathematics.
Counterpoint: but couldn’t they simply do a bunch of lower-priority tasks, whereupon anytime someone needs something from them they can easily drop that and shift over to do that at a higher prioritization? Yeah it’s wasteful for context switching, but it gets the main job done and that’s what matters?
ELI5 version: every new request from an actual human goes straight to the front of the line, or rather to the back of the “human” line, in front of all the “busywork”.
Yes, as long as you accept that the lower-priority tasks get dumped when needed.
This is a common way to deal with it. But the number of managers that know how to decide a task is low-priority is exceedingly small. Most only have top-priority tasks to distribute to people.
Though the absolute best cluster systems I’ve seen have utilized this principle correctly, never leaving it idle, yet never blocking work that others want to do either (for more than a very small amount of time).
Planning such takes a great deal of effort though, and most people seem to simply want to be paid and even more importantly than that feel in control, or perhaps worry that if they don’t rise up beyond their potential to handle matters that their own job won’t be quite as stable. Bc capitalism seems to fuck up everything it touches, more’s the pity.:-(
Any manager that doesn’t know about the utilization/latency trade-off from queue theory is a danger to themselves and to others.
ELI5?
If you have somebody doing work that can appear at random (like somebody calling and saying they have a problem), that person will either be free for a fraction of time that seem high to naive people, or will have a line and take ages to help anybody approaching them.
That seemingly high fraction of time is usually around 50% for the line to stay under control. That’s a well known result from mathematics.
Counterpoint: but couldn’t they simply do a bunch of lower-priority tasks, whereupon anytime someone needs something from them they can easily drop that and shift over to do that at a higher prioritization? Yeah it’s wasteful for context switching, but it gets the main job done and that’s what matters?
ELI5 version: every new request from an actual human goes straight to the front of the line, or rather to the back of the “human” line, in front of all the “busywork”.
Yes, as long as you accept that the lower-priority tasks get dumped when needed.
This is a common way to deal with it. But the number of managers that know how to decide a task is low-priority is exceedingly small. Most only have top-priority tasks to distribute to people.
Sigh… yes.
Though the absolute best cluster systems I’ve seen have utilized this principle correctly, never leaving it idle, yet never blocking work that others want to do either (for more than a very small amount of time).
Planning such takes a great deal of effort though, and most people seem to simply want to be paid and even more importantly than that feel in control, or perhaps worry that if they don’t rise up beyond their potential to handle matters that their own job won’t be quite as stable. Bc capitalism seems to fuck up everything it touches, more’s the pity.:-(
- image source
Q theory, you say?
I think a lot of them don’t know their ass from a hole in the ground.