VEGAS Pulsar Project Meeting: 2015 November 23rd 2:00 - 3:00pm ET

Room / Connection Details

  • GB-137 / Soc 280
  • 192.33.117.12##7144
  • 434-817-6443

Housekeeping Details

  • Main wiki page
  • gbsapp Mailing List
  • Location and regular date and time of meeting - agreed weekly on Mondays; GB-137 / Soc-280; 2-3pm ET
  • ETK codes:
    • Electronics: 432510.GB3015
    • Software: 893252.GB3015
    • Science: 432540.GB3015

Resource Allocations

BLOSS, MARTIN ALAN 0.05
BRANDT, JOSEPH JOHN 0.10
CREAGER, RAY 0.65
DEMOREST, PAUL 0.05
RICHMOND-DECKER, JUSTIN 0.20
FORD, JOHN M 0.05
MELLO, MELINDA JANE 0.05
MCCULLOUGH, RANDY L 0.25
RAY, JASON 0.55
PRESTAGE, RICHARD 0.05
LYNCH, Ryan 0.25
VAN TONDER, VEREESE 0.10

Present

  • Richard, Marty, Vereese, Justin, Ryan, Ray, Paul

BOF file status updates

  • c1500x4096 mode
    • John installed new GTX 980 and Justin updated guppi_daq on srbs-hpc1
    • Ryan tried testing this mode using a normal scan but is still seeing packet loss
    • Joe, Ray, John, and Ryan discussed possible causes and next steps
      • Put guppi_daq in MONITOR and look for packet loss: DONE - No significant loss seen
      • Check to see which child process is using the most resources in both original guppi_daq and gpu_transpose versions (TODO)
      • Check NETSTAT for blocked message: DONE - In at least some scans it does start out in Blocked, but then starts running as expected with packet loss
        • Paul suggests checking if this happens at the start of the first scan or at the start of every scan
        • Shared memory buffers could be locked
        • GPU intiitalization step can throw things off at the stop of a scan
      • Use nvidia profiler to check CUDA processes (TODO)
    • John, Paul point out that the data rate depends only on bandwidth, not on of number of channels
  • c1500x0512 mode
    • BL may want to use this mode since it matches SETI@home
    • Ryan tested on srbs-hpc1 with new GPU. No significant packet loss seen, and AP shows up nicely. Ryan will update the wiki page with these results and will test other coherent modes where we were seeing dropped packets.
  • Fix for nchan calculation in coherent modes
    • Both bugs are identified and solutions are in hte c1500x4096 mode, ready for propogation to other designs
  • Bandpass issues in i0800x0064, 0128, 8192
  • Duplicated signal in i0800x8192
  • Use low bandwidth spectral line modes as a basis for the 100 MHz and 200 MHz pulsar mode bofs
    • Will require development on Randy's part to work out filter coefficients
    • Currently a time series is output rather than a spectrum
      • Would need to transform this and apply dedispersion filters
    • Would need to implement PFB code if it is not done on the GPU
      • Might be able to use PFB code from VEGAS
    • Could single GPU keep up with coherent dedispersion?

Software

  • Configtool MR
    • Justin got got config tool to output a correct list of parameter settings
      • Currently have to configure rest frequencies using dictionary structure, otherwise config tool tries to put all "windows" in the same bank
      • Justin will double check that config tool does not think it is configuring for a multi-subband mode (emulate VEGAS spectral modes 1--19 in this regard)
  • GBT Status shared memory parameters in PSRFITS headers
    • Justin located the python code that updates shared memory from GBT status database.
      • In GUPPI runs through a supervisor daemon
    • Justin wrote a C++ class that emulates this behavior and incorporated into manager
      • Only runs in pulsar
    • Ryan tested in simulator and the PSRFITS headers seem correct
      • Will test with Device Explorer to set observer and source, and then with GBT
  • Switching signal generation
    • Justin has made changes to the Manager that should invoke the relevant modules from the backend class
      • Ryan tested with GBT hardware but VEGAS was not switching properly. Needs more investigation.
    • Randy discovered that there are two values of nchan in the designs, one for switching signal generation and one for the rest of the logic
      • Need to test on srbs-hpc1
  • Matrix HPC
    • No updates

GBT testing

  • c1500x4096 mode has same dropped packet issues using VEGAS hardware.
  • Ryan will focus on vetting the c1500x0512 mode with VEGAS hardware to support BL.

Wideband

  • No updates

VEGAS/DIBAS-style backend for FAST

  • Budget to be integrated back into write up
  • Marty suggests sending to Karen and if she gives approval, submit new ideas form
  • Richard wonders if it would be better to have FAST ask NRAO for a first-light backend
  • Action item for everyone: Give current proposal another read for consistency and correctness. Finish by one week from today.

Updates on new BL switch installation

  • New switch installed with new fibers
    • 5 Tb/s theoretical maximum capacity
    • Marty asks if we have a spare
      • John recommends keeping old switch as a back up, but notes that these are usually reliable

AOB

  • John is interested in using multicasting for both the spectral line and pulsar modes
    • Would require some modifications to guppi_daq and VEGAS modes
    • Also require updating ROACH config
    • Switch should be able to handle all other routing as long as ROACH+HPC 10 gig port is a separate multicast group
    • Would allow us to run VPM during BL observations, for example

Date / Time of Next Meeting

  • November 16

-- RyanLynch - 2015-11-23
Topic revision: r2 - 2015-11-23, RyanLynch
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding NRAO Public Wiki? Send feedback