You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I'm running in bluefors with a reported 2x64 (actually should be 2x32 I think) abaco source with a reported data rate of 32.3MB/s and 4.096 us/sample. I'm running edge multi triggers and off files and top reports 250-300% cpu usage for dastard.
This compares to about 200% cpu usage for a tdm system of 8x32 at 200 MB/s, which should be doing more work than the above config. Likely culprits include parsing all the packet headers and biased unwrapping.
The text was updated successfully, but these errors were encountered:
We'll plan to do a comparison head to head between µMUX with Abaco+UDP vs a TDM system at comparable data rates. (If a TDM has 200 MB/sec, but only half that is from feedback channels, is a "comparable" µMUX system running at 100 or 200 MB/sec?)
Things to investigate:
How does the load reported by top change when you change the # of channels? How about the profile reported in the go profiling tool?
I think that the AbacoSource.readerMainLoop() shows that we're doing quite a few activities in a single goroutine. The most important might be the group.demuxData(...) call, which we've seen to cause some 10-20% of all CPU time (according to the profiler, at least). If necessary, maybe we can think up a way to parallelize the demux step, sharing among maybe one goroutine per channel group?
EMT triggering algorithms can probably be made more compute-efficient. (But we're not there yet!)
I'm running in bluefors with a reported 2x64 (actually should be 2x32 I think) abaco source with a reported data rate of 32.3MB/s and 4.096 us/sample. I'm running edge multi triggers and off files and top reports 250-300% cpu usage for dastard.
This compares to about 200% cpu usage for a tdm system of 8x32 at 200 MB/s, which should be doing more work than the above config. Likely culprits include parsing all the packet headers and biased unwrapping.
The text was updated successfully, but these errors were encountered: