Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.atnf.csiro.au/vlbi/dokuwiki/lib/exe/fetch.php/difx/meetings/sydney2012/sydney2012-kettenis-data.pdf Дата изменения: Mon Sep 24 07:31:26 2012 Дата индексирования: Tue Apr 12 12:46:23 2016 Кодировка: Поисковые слова: спрайты

Handling Bad Input Data
Mark Kettenis JIVE


Correlating Bad Data
Goals: · Recover as much data as possible · Minimize the amount of bad data reaching the PI We all hate flagging! · Don't slow down correlation In SFXC we: · Keeps track of blocks of bad data · Set samples in these blocks to 0.0 during conversion to floatingpoint Same mechanism is used to handle data-replacing headers (Mark4)

DiFX meeting ­ Sydney - 24 September 2012Date


The Weakest Link?
The Mark5 data recorder problems: · Not all data makes it onto the disks · Disks failures
Data is replaced by fill pattern

SFXC searches for fill-pattern in input data
Optimized by recognizing fill-pattern blocks are 64k

An annoying problem remains: · XLRRead sometimes fails instead of return fill pattern
Further XLRReads fail as well, even at different byte address Only a full reset of StreamStor hardware fixes this

Problems are disproportionally bad at 1 Gbit/s
DiFX meeting ­ Sydney - 24 September 2012Date


Mark5A issues
· Mark4 Formatters develop bad tracks Complicates looking for sync words Bad track means we have to throw away entire subbands Seen something similar with VLBA recently
But only once so far

· Incomplete Mark4 frames Easy to detect: sync word not at expected location Expensive to fix: resync data stream
SFXC uses Horspool's algorithm

DiFX meeting ­ Sydney - 24 September 2012Date


Mark5B Frame# Corruption

· Frame# supposed to start at 0 at second boundary Sometimes just keeps increasing Sub-second VLBA timestamp still correct · SFXC switches to using VLBA timestamps if needed Sub-second VLBA timestamp not always accurate enough
Use additional heuristic to "fix" timestamps

DiFX meeting ­ Sydney - 24 September 2012Date


Questions?
Research leading to these results has received funding from the European Union's Seventh Framework Programme (FP7/2007 2013) under grant agreement n° RI-261525 NEXPReS. This material reflects only the author's views, and the European Community is not liable for any use that may be made of the information contained therein.

DiFX meeting ­ Sydney - 24 September 2012Date