�� . �� : http://www.naic.edu/~phil/pdevall.html
�� : Tue Dec 10 20:20:58 2013
�� : Thu Feb 27 20:52:05 2014
��:
�� : http www.astronomy.ru forum index.php topic 4644.0.html

datataking_software

Jeff mock's pdev spectrometer

03jun08

History:

See also:

Documentation:	Hardware	Jeff's pdev users guide (.pdf) pdev hardware: description, specs, and measurements. Hardware Diagnostics Maintenance procedures Places to look for debug info
	Software	pnetctl: controlling pdev from cima File descriptions: ao fits file format for pdev spectrometer psrfits file format for pdev spectrometer .pdev file format pdev software: jeff mock's and ao additions Radar datataking with pdev
	Misc	areca raid controller (1220) manual (.pdf) idl routines to process the .pdev and .fits files Galfacts info. Notes:
Measurements	All Measurements in Chronological order Datataking measurements Galfacts info. Monitor pdev hardware (fans, temps, voltages) Psrfits test data.
Problems	Problems that have occurred.

History:

131024: dUpdated software on pdevs0 to most recent code. Switched the working directory on pdevs0 to use the http server ( it was still using the megs2 repository). Decided to keep the box names b0spN rather than b0sMgN .. Will cause a little hassle moving config files between pdevsN and pdevs0.
131007: mock synth's have unlocked error. Dana measured jumps in 10 mhz reference (at output of 3 synths 10 mhz daisy chain more info) . Jumps of up to a few cycle. About 6 in a few days. We saw this a few years ago. The synth with the 10 mhz from buffers never has jump (just the daisy chained ones). The synth's re regenerating the 10 Mhz.
121105

replaced pdev-108 (b1sXg1) with pdev-101.
We were having lots of bit flips in the data b1s1g1 polA. polB looked ok.
redid cal.conf for new pdev-101 a/d

120221

replaced pdevs7 spectrometer (pdev-109) with pdev-127. We were having lots of dump timeouts. Dmesg on pdev-109 showed some ethernet 0 timeouts.

110919

replaced b5sxg0 box 104 with box 112. Chris had problems with jumps in b5s1g0

The jumps in the spectra looked like bit flips in the data.
when running adjpower, there were many a/d overflows (looking at /tmp/pdevsig.log on pdevs1). If the bit flips were occurring, they must have been occuring vary early in the pipeline since the a/d overflow bit check is right after the a/d and mux switch. This bit then gets carried along with the data sample. Since the bit and the data sample were probably bad, it really must have been a problem at the early stages.
Ran diagnostics before box 104 replaced.

ltest,rtest ok
rdcomp failed. mainly with b5s1g0. (b1s1g1 also failed a lot).

1106xx:

11jun11: pdevs6 disc? in raid fails. replaced
17jun11: pdevs2 disc2 in raid fails. replaced

100908

pdev-113, pdevs7, b6sxg0 fan2 (gxb) dropouts). looks like power supply trouble
pdev-113 replaced with pdev-109

100902:

pdev-107, pdevs3,b2sxg0 fan 2 replaced.

090906:

b1sxg0 boxes fan failed, fpga died. This was pdev-109. Replaced with pdev-103.

090214:

Swapped I,Q cable in spectrometer box: b6s1g1 (bm 6, high band, polA): dig i0,q0 signal cables swapped inside spectrometer box. Probably occurred when the dig levels were changed from 1.5 to .78 volts. Group1 has not yet been used for datataking. 090216 check all the other boxes for i/q switches.. None other found.

081205: replaced disc 5 on pdevs1 raid.

The disc i/o had not been keeping up with galfacts datataking (occasionally).
The raid stats showed that disc 5 had more seek errors, timeout cnts..

081010

Accidentally overwrote pnetctl.html.. Grabbed backup from 12sep08.. and then fill in some of the lost stuff.
B2SxG1 switched out spectrometer box pdev-105 . replaced with pdev-122. connects to pdevS10 server. digitizer S0 IA was giving problems.

081001:

Used a test program to write as fast as possible to a psrfits file (input came from memory) (.ps) (.pdf).

080821: svn checkout https://mock.com/svn/pdev/web web on pdevs1:/home.local/pdev/pdev to get the web files for the website.
080615: timedomain sampling.

Does not have an 8 bytes hdr rec. psrv.c was treating those bytes as a rec header and byteswapping them.
psrv.c was also using the fftlen to compute how many blks fit in 1 second. Updated so it used the sample rate.
pnetbld now correctly computes how long each block takes in time domain mode.

080614: polA,polB or stokesI is stored as unsigned int while fits uses signed. Changed bpOut.c to clip to maximum signed value for these pols.
080613: ashift computation in pnetbld. Shift up by 1 more for stokesU,V relative to pola,polB. This makes the rms of U,V the same as polA,B since jeff outputs 2*polA,2*polb, but only 1*U, 1*V. (Note: 13jun08 i shifted 1 less for u,v..17jun08 i corrected it to shift 1 more). check data
080608: pnetbld changed default hr_shift to be 12-alog2(sqrt(dec)). see fpga dlpf and parameters hr_shift. check
080415:attenuation before a/d changed by 6db.. went from 1.5 to .78 v to get 2048 counts.

NOTES:

These notes need to eventually be added to the documentation above:

0910220: pnet 1 second messages output to the user

36.49 MB/s 20.68 MB [1:1]/1000 blocks (0.1%)
Iorate:

This is a cumulative (in time) i/o rate from the start of the observation.
Computed in the prun spectrometer (prun_rmtdma()). Output by pnet.
The i/o rates from all of the spectrometers are summed. debug mode probably lets you see individual rates.


home_~phil