Currently the depth queue is hardcoded to 50... however one needs to keep #Threads x Depth below a certain number otherwise fetcher spends its life in managing the queues and CPU becomes the limiting factor. This is what was creating my L shape kind of bandwidth usage.
I had to patch it by hand, however I suggest we create a conf item to enable different combination #Threads x Depth. Or we could also set it the opposite way which is to define a "max.fetchqueue.totalSize" and derive the depth as totalSize/#Threads. And of course max.fetchqueue.totalSize is very dependent on the system specs, but this is what you want to control. -- -MilleBii-
