Experiences with PerfSONAR and a Control Plane for Software Defined Measurement Yan Luo Department of Electrical and Computer Engineering University of Massachusetts Lowell Slide 1 Slide 1 FLowell Project at a Glance • Legend Campus Backbone Network Core Router/ Switch Science Network (10G) Science DMZ Non Science Network SDN OpenFlow Switch OpenFlow Controller Internet ESnet PerfSONAR Academic Buildings OpenFlow OpenFlow Switch Switch research research labs labs Border Router (*.97) UMassNet Internet2 Firewall Firewall Border Router (*.219) UMASS LOWELL CAMPUS Science DMZ Core Router/ Switch SDN OpenFlow Switch FLowell Rack Other collocated racks MGHPCC Slide 2 FLowell Project Status • UML Campus SDN Network – In-lab Testbed Completed: Six OpenFlow switches (Extreme Networks) + OpenDayLight Controller – Campus-wide Deployment In Progress (two buildings completed, other two buildings by Oct’15) • Layer-2 10Gbps Link (UML<--->MGHPCC) in progress (new Ciena optics pack in order) • Layer-2 10Gbps Link (UML<-->Internet2) in progress (new Cisco switch ordered) • Science DMZ in progress (architectural design and verification, vendor selection) • GENI Rack in progress (vendor selection) Slide 3 Performance Tests • DTN – CPU: Intel Core 2 @2.33GHz, 2 cores – Hard Disk: Read/Write Speed: 125MB/s – 1Gbps edge link (production net) • PerfSONAR Inside firewall File Size Speed Ave(MB/s) 10M 3.62 50M 11.92 100M 23.84 1G 42.47 10G 37.48 50G 36.51 Test server: anl-diskpt1.es.net Outside firewall Slide 4 PerfSONAR Today • Over 1400 public perfSONAR nodes Slide 5 PerfSONAR’s Scaling Challenges • Challenges – Control, coordination and execution of network measurements – Monitor healthiness of networks besides major networks – Network issues caused by multiple problematic links • Our Research – Using measurement archives (MA) to build a traceroute graph – Propose a control plane on top of perfSONAR to support software defined measurement and troubleshooting – A joint work with ESnet (Brian Tierney) and AMPATH/FIU (Jeronimo Bezerra) Slide 6 Motivation of PerfSONAR Control Plane • Typical Workflow • Finding the Longest Clean Path Slide 7 Objectives of PerfSONAR Control Plane • Measurement Archive Data Analysis – How were the measurement results? – What can we learn from them? • Automatic perfSONAR Peer Selection – Quickly identify the best suitable PS node(s) on the routes in question • Programmable Measurement and Troubleshooting – Define measurement task and conditions with software Slide 8 The Design of PerfSONAR Control Plane • PerfSONAR Node Discovery – Finding nearest perfSONAR node of a target router on the path • Measurement Task Control – Initiating tests between (any) two chosen perfSONAR nodes – Monitoring the performance on the path – Locating the problematic link(s) Slide 9 The Operation of PerfSONAR Control Plane • Obtain traceroute information from MAs • Build a traceroute graph based on the dataset • Find a set of perfSONAR node pairs to start bandwidth tests and monitor the results • Diagnostic analysis and troubleshooting network issues Slide 10 Evaluation of PerfSONAR Control Plane • Traceroute Dataset – 95 MA hosts in the central US and eastern US regions – 1831 traceroute records • Traceroute Graph – 2377 perfSONAR hosts and routers in total Slide 11 A Use Case of the New Control Plane • pr20.uml.edu --- typhoon.pub.alcf.anl.gov (140.221.68.2) • A python program with less than 300 lines of code • “Troubleshooting” procedure took about 15 minutes Slide 12 Conclusion and Future Work • FLowell project at UMass Lowell in good progress • Gained experiences with DTN and PerfSONAR • A Control Plane for PerfSONAR show promising results – Open source at: https://github.com/ACANETS/pscp – Community feedback welcomed! • Tasks in the Upcoming FY: – – – – Science DMZ Deployment at UML GENI Rack operational at MGHPCC Campus wide outreach to researchers PerfSONAR Control Plane Collaboration Contact info: Yan Luo, yan_luo@uml.edu Slide 13