Data Transfer Nodes (DTN) configured in La Serena and Champaign with nuttcp (a network performance measurement tool) generated a sustained memory-to-memory data rate over 80 Gb/s, over a period of three hours. Simultaneously, the DTNs, using the Fermilab Multicore-Aware Data Transfer Middleware (MDTM) software, achieved a peak of 36 Gb/s transferring 200 Gigabytes of DECam public data (FITS files) provided by the National Optical Astronomy Observatory (NOAO). Note that in LSST operations, there will be over 20 DTNs (aka archiver/forwarders) simultaneously sending data, so each one will require far less than 36 Gb/s. In addition, on the Champaign end the files were ingested into a GPFS shared file system, and a Jupyter Notebook running an application provided by LSST Data Management was used to visualize the files. Finally, an additional test transfer from Champaign to La Serena is being conducted and has so far achieved a peak of 40 Gb/s, sufficient for the annual transfer of LSST Data Releases to Chile.
Instrumentation in the DTNs and links and Grafana software were used to provide a real-time, web display of network performance during the demonstration. This was monitored live from the NCSA booth at the Supercomputing 2018 Conference. A number of conference attendees witnessed the demonstration and presentation, and participated in a question and answer session.
According to LSST NET Lead Jeff Kantor, “This demonstration shows not only that we have continuity and performance from the network point of view, but also that all of the partners acted as a very well-coordinated engineering team for LSST.”
Congratulations to the LSST NET SC18 Demonstration Team:
Albert Astudillo (REUNA)
Jeronimo Bezerra (FIU/AmLight)
Julio Ibarra (FIU/AmLight)
Sandra Jaque (REUNA)
Matt Kollross (UIUC/NCSA)
Ron Lambert (LSST/AURA)
Sean McManus (NOAO)
Wil O’Mullane (LSST/AURA)
Rodrigo Pescador (RNP)
Andres Villalobos (LSST/AURA)
Adil Zahir (FIU/AmLight)
Additional support was provided by a number of other people within these organizations. We are particularly grateful to SCInet, Starlight, and Fermilab for enabling this demonstration.
Original LSST article published here