VEGAS Hardware Tests - 2015 Sep 22
- Investigate source of incorrect sub-integration times on certain banks
- Investigate switching signal control with VEGAS in pulsar modes
We attempted to run scans in various coherent modes in order to investigate why some sub-ints have incorrect values of tsub (aka tfold). The tfold parameter was being set properly in shared memory (typically to 10 s) but the PSRFITS headers recorded values less than this, and the exact values varied from bank to bank. In particular
- In c0800x0064 mode banks A, G, and H consistently record one extra subint, and the first subint has a duration of 20.48 us. All other subints have the correct duration. All other banks perform as expected.
- In c0800x0128 all banks have subint durations of 3--5 s.
- The c0800x0128 mode had significant packet loss.
We initially thought the packet loss was a result of the incorrect subint durations, but we later realized it was the cause. The PSRFITS headers record the subint duration based upon the number of samples that are actually recorded. The 50-70% packet loss corresponds with the calculated subint duration. Ray noted that GUPPI required reads a configuration file (guppi_threads.conf) that contains bitmasks to pin certain guppi_daq processes to certain cores. The file being used by VEGAS is not optimized.
We also looked into the switching signal generation. The requested switching signal is correct but the actual switching signal is not.
Ray and Joe will suggest proper bitmasks for various guppi_daq processes. Tests will be repeated with this new configuration. Further investigation of the switching signal control is needed.