I have made a bunch of tests, and using blocking IO is not the answer. Since (as per Andreas) multi-source transfers easily reach the line speed, this is not really a priority for the moment. I will look into this further after we're done with the release candidate for 4.0.0.
Non-blocking, multiple streams:
[root@p05153074285630 tmp]# time xrdcopy --streams 6 -f root://p05153074366867//blah /dev/null
[29.3GB/29.3GB][100%][==================================================][666.7MB/s]
real 0m45.483s
user 0m0.922s
sys 0m43.342s
Blocking, multiple streams:
[root@p05153074285630 tmp]# time xrdcopy --streams 6 -f root://p05153074366867//blah /dev/null
[29.3GB/29.3GB][100%][==================================================][447.8MB/s]
real 1m7.180s
user 0m0.193s
sys 0m46.521s
Non-blocking, single stream:
[root@p05153074285630 tmp]# time xrdcopy -f root://p05153074366867//blah /dev/null
[29.3GB/29.3GB][100%][==================================================][500MB/s]
real 1m0.826s
user 0m0.470s
sys 0m52.185s
Blocking, single stream:
[root@p05153074285630 tmp]# time xrdcopy -f root://p05153074366867//blah /dev/null
[29.3GB/29.3GB][100%][==================================================][468.8MB/s]
real 1m4.745s
user 0m0.179s
sys 0m46.570s
As you can see, blocking IO gives worse real time in all cases, and negligibly better system time for the single stream example. Note: this is roughly an average of multiple tests for each case.
—
Reply to this email directly or view it on GitHub.
Use REPLY-ALL to reply to list
To unsubscribe from the XROOTD-DEV list, click the following link:
https://listserv.slac.stanford.edu/cgi-bin/wa?SUBED1=XROOTD-DEV&A=1