• IP adress
  • GPU server
    1Gb port: 159.226.170.68
    10Gb port: 192.168.1.127 # p2p1, FRB power term
    10Gb port: 192.168.2.127 # p2p2, FRB cross term
    10Gb port: 192.168.3.127 # p2p3, SETI pol 1 term
    10Gb port: 192.168.4.127 # p2p4, SETI pol 2 term
    
  • roach2
    1Gb port: 10.128.2.8
    10Gb port: 192.168.1.227 # FRB power term
    10Gb port: 192.168.2.227 # FRB cross term
    10Gb port: 192.168.3.227 # SETI pol 1 term
    10Gb port: 192.168.4.227 # SETI pol 2 term
    
• 10Gb network tricks in Hashpipe
  • Code using packet sockets (or raw sockets) will see packets before iptables sees them. The s6_pktsock_thread uses packet sockets to receive the packets. The IP stack is not aware of this, so as far as it is concerned there are no listeners so for each S6 packet received the IP stack would send a "no listener" ICMP packet in response. These ICMP packets were generating unwanted network traffic which impacted the overall throughput so to avoid them we setup iptables rules to drop all incoming UDP packets (since the only UDP packets expected were S6 packets).These iptables rules stopped the ICMP packets, but also stopped "normal" UDP receive code from seeing the packets.
  • If you want to use normal UDP receive code, you need to remove these rules. This can be done via:

   sudo iptables -F -t raw

  • You can check for the rules before and after running that command by using this command:

   sudo iptables -L -v -t raw

  • If some of the interfaces on a netbooted machine do not come up, issue a:
    

   sudo  ifup -a

• roach2_conf.py is used to load the boffile to roach2, setup ip address and configure parameters.
• adapt_gain_conf.py is used to find a proper gain value for 8 bit output data. Set the gain value step by step to increase or decrease the value of output data. Once the average data falls into the preset range, the entire adaptive adjustment is completed.
• receive_packets.py is used to receive the packets from all four 10Gb ports and print the header information out.
• netspeed.sh is used to test the network speed of a given IP adress.
• topology_pnn.sh is used to get the NUMA information for Hashpipe.
• psrsoft_cmd has some command can be used for pulsar data processing which form different software such as sigproc,dspsr,presto,psrchieve,heimdall ...
• heimdall_script is shell script run on computing node for filterbank file looking and heimdall launching.
• monitor_10gbe.py is used to monitor the output packets from 10GbE.Receiving the packets from 10GbE and print the head out and plot the spectrum data out.
• waterfall_plot3D.py is used for 3 dimensions waterfall figure plot out. Receiving a given number of spectrums from 10GbE and plot in frequency, time and power.

• Generate a fake periodic pulsar signal from a sweep generator,merge the signal with a noise and inject to both polarizations.
  • sweep frequency: 500MHz to 100MHz
  • period: ~1.5 sec
  • signal amplitude: -26dBm
  • attenuation of noise source: 20dB
  • DM = (t2-t1)/(4.15ms)[(ν1/GHz)^-2-(ν2/GHz)^-2]^-1=1500/4.15/(1/1.4²-1/1.8²)=1793
• The spectrum plots of both polarizations are shown in Fig.1

Fig. 1 spectrum of both pols

• 3D waterfall plot from 10Gb port is shown in Fig.2

Fig. 2 3D waterfall plot from 10Gb

• Waterfall pictures before and after dedispersion by DSPSR for filterbank file are shown in Fig.3 and Fig.4
• 

Fig. 3 Waterfall before dedispersion Fig. 4 Waterfall after dedispersion

• Time domain plots before and after dedispersion by DSPSR for filterbank file are shown in Fig.5 and Fig.6
• 

Fig. 5 time domain before dedispersion Fig. 6 time domain after dedispersion

Use heimdall to search filterbank file which sweep generator faked with following command:

time heimdall -f B0_I_2018-02-09_13-24-53.fil -dm 1000 2000 -nsamps_gulp 100000 -rfi_tol 50 -boxcar_max 16 -dm_tol 1.5 -v

What heimdall found in a cand file as follows:

• Generate a faked pulsar signal and noise signal from sweep generator and noise source seperately, merge signal with noise to pol1 and only inject noise to pol2, the setting of the sweep generator is the same with single beam test.
• In Hashpipe data recording, we can select which beam you want signal injecting and which beam you wouldn't. For example, odd number of beams have signal and even number of beams only have the noise.
• The waterfall pictures have processed by DSPSR of those beams with or without signal are shown in Fig. 7 and 8.

Fig. 7 dispersion result of signal and noise Fig. 8 dispersion result of without signal