This page has examples of NetLogger analysis.

GridFTP servers at SC09

For the SC09 Bandwidth Challenge, NetLogger monitored GridFTP servers at LLNL, NERSC, and ALCF. The graph below shows throughput from LLNL. The NetLogger analysis code summed overlapping transfers from each server (colored lines) and also summed all three servers to arrive at an estimate of overall throughput. 

The graph shows the throughput from three servers at LLNL to the SC09 show floor. The colored lines are individual servers, and the grey line is the total. The server in red (gdo1) had disk problems which limited its throughput.

For more information, see:

Data transfer throughput for small files

Anecdotal reports described low throughput for STAR scientists transferring files between Brookhaven National Laboratory (BNL) and NERSC PDSF in Oakland, CA. Some investigation of the GridFTP logs showed that the nature of the transfers -- a mix of small and large file sizes -- caused sub-optimal performance for the globus-url-copy command, which was being invoked per-file

 The graph at right shows all file transfers between hosts at BNL (ending in bnl.gov) and the GridFTP server at PDSF (pdsfgrid3.lbl.gov), plotted with file-size in bytes versus achieved bandwidth in Mb/s.

to be continued..

Workflow performance profiling