Follow-up Comment #1, bug #87560 (project xrootd):
More info:
1) Andy believes there are some code paths (when sendfile is used) where
statistics aren't kept for reads.
2) ReadV's aren't logged currently (likely is the cause of the majority of
our issues, as CMS jobs should be over 99% ReadV).
For (2), our preference would be to have a separate trace format which
includes the number of bytes, the number of chunks, and *maybe* the first
offset (we don't feel strongly about the offset). I don't think it'd be
useful to actually unroll and list out all the reads individually.
_______________________________________________________
Reply to this item at:
<http://savannah.cern.ch/bugs/?87560>
_______________________________________________
Message sent via/by LCG Savannah
http://savannah.cern.ch/
