LRB window tuning about lrb HOT 4 CLOSED

Hi @lynnliu030 ,

Sorry for the late response. Your code runs correctly. I updated the doc and you can find the explanation of the output here: https://github.com/sunnyszy/lrb/blob/master/LRB_WINDOW_TUNING.md.

I also updated the Github sample config. Here your current version would be slow because it only uses one concurrent process in simulating the 150 tasks sequentially. To make it faster, you can pull the latest Github version and I use GNU parallel to automatically config the amount of parallelism. You can also manually adjust the parallelism by changing the nodes in https://github.com/sunnyszy/lrb/blob/master/config/job_dev.yaml#L9 from localhost to x/localhost, where x is the number of parallel runs you would like.

from lrb.

@sunnyszy Sorry, but one more question, how long does it normally take for the simulation in the parallel mode? Say for the wiki-18 or -19 trace

from lrb.

Hey @lynnliu030 ,

Each simulation is 20% of the time of the full trace. Depends on your # threads, you should be able to calculate the time. Also, when the algorithm decides to use smaller cache sizes to do linear fit, it needs additional rounds of simulations.

from lrb.

@sunnyszy Thanks for the info!

from lrb.