Closed mpokorny closed 6 years ago
It seems that the generation of random normal deviates is kind of slow! Just looking at "top" on the node running vyssim shows that one of its threads reaches 100% cpu usage fairly quickly as the number of antennas increases. Since vyssim synthesizes timestamps simply based on a starting time and a count of the number of frames (at the expected rate), its timestamps tend to fall behind when vyssim can't keep up in real time. One possible workaround would be to run vyssim on several nodes in parallel (which it is designed to do, but hasn't been tried much beyond some early testing of mine). In the meanwhile, I'll have a look at the rng and see if there's anything I might to do speed it up. However, I don't imagine I'll easily come up with anything close to the roughly order of magnitude improvement that would be needed to go from 8 to 27 antennas in my test case (16 spw, 32 ch/spw, 4 pol), so using vyssim in parallel mode will likely remain a necessity in some cases.
Closing this issue, as the alternative of running vyssim in parallel is certainly a reasonable option, and there is not an easy way to speed up the normal rnd generation.
Timestamps coming out of vyssim when it is generating normal random deviates for visibilities fall behind the system clock as vyssim runs.