perfSONAR User Q&A and Other Discussion

["Andrew Lake" <>]

All,

A patch for the perfSONAR Toolkit has been released that fixes a long-standing bug in how the TCP congestion control algorithm is set. The bug is that the TCP congestion control algorithm is always set to “reno”, regardless of the state of NDT or NPAD. The intended behavior is that if NDT or NPAD are enabled, the algorithm is set to “reno”, otherwise it is set to “htcp”. In general, performance will be significantly worse with reno, but is required for NDT and NPAD since they assume specific reno behavior in their heuristics.

This has been corrected, so those of you not running NDT or NPAD may notice a significant increase in your throughput results once you install the patch. Please note that this bug was NOT introduced in the latest 3.5 release and dates back at least to the 3.4 release. For those of you still running either NDT/NPAD it may be a good time to review your use of these tools and whether enabling them is worth the associated performance hit. We strongly suggest not running NDT/NPAD on the perfSONAR host that you use for regular testing, and recommend using a separate, dedicated host for NDT instead.

For those running auto-updates, you should have the patch shortly. To manually update run the following:

yum update perl-perfSONAR_PS-Toolkit*

You will know if you have the patch if the toolkit version on your web interface is set to 3.5.0.5. Please let us know if you have any further questions.

Thank you,

The perfSONAR Team