Network Performance Measurement Atlas Tier 2 Meeting at BNL December 14 2005 Joe Metzger email@example.com
Network PerformanceMeasurement • Motivation • Requirements • Measurements • Useful tools • Suggestions
Motivation • LHC science is going to rely on a very complex web of many different networks. • Networks are not static. • They are in a constant state of change. • Components fail. • Wide area links are affected by plane crashes, train derailments, fires, floods construction, etc.
10 Gigabit networking will present new challenges. • The high costs of 10GE routers is pushing people to deploy different architectures using lower cost switches. • Many of these lower cost devices provide very limited diagnostic and debugging support. • Over subscription of circuits and bottlenecks within equipment may cause un-expected packet loss. • The differences between OC192, 10GE LANPHY, 10GE WANPHY and almost 10GE line-rate equipment may cause buffering problems. • These issues will lead to problems that will be challenging to identify and resolve.
Requirements • You must have the ability to easily determine the status of the networks you rely on for your missions. • Up and working correctly? • Can you prove it? • Down • Is there a known problem that is being worked on? • Are you seeing a symptom of the problem or something else? • Is the network down or the applications down? • Can you prove the problem is not on your campus? • Who do you call? • What info can you provide to the NOC who will fix it? • Up but not performing as expected. • Is there a known problem? • Who do you call? • What info can you provide to help identify the problem?
Measurement • Paths • What set of links do your packets traverse? • Capacity • What is the capacity of those links? • What is the utilization of those links? • Bandwidth • What is the achievable bandwidth across the path? • Is it stable or does it change? • Latency between points of interest • What is the round trip time across the path? • Is it stable or does it change?
Tools • Paths • Traceroute • Latency • Ping and OWAMP • Utilization • Collect with MRTG/Cricket/SNAPP etc • Publish with PerfSONAR. • Bandwidth • Iperf, netperf, nuttcp and BWCTL
Traceroute ESnet PerfSONAR Traceroute Visualizer Trace Submitted Tracing route to cache3.bnl.gov 188.8.131.52 over a maximum of 30 hops 1 1 ms 1 ms 1 ms joem-fe-stub.es.net 184.108.40.206 2 21 ms 21 ms 21 ms chi-ameslab.es.net 220.127.116.11 3 21 ms 21 ms 21 ms chicr1-ge0-chirt1.es.net 18.104.22.168 4 41 ms 41 ms 41 ms aoacr1-oc192-chicr1.es.net 22.214.171.124 5 43 ms 43 ms 43 ms bnl-oc48-aoacr1.es.net 126.96.36.199 6 43 ms 43 ms 43 ms bnl-esbnl.es.net 188.8.131.52 Analyzing Trace
Recommendations • Deploy test servers at the edge of your network. • OWAMP for measuring latency. • IPERF/BWCTL for measuring bandwidth. • Monitor interface utilization • Capture with MRTG, Cricket, SNAPP or other tools. • Export results to community using a PerfSONAR Measurement Archive.
Summary • Deploy tools that will allow you to: • Determine if your applications are working correctly. • Determine if your network is working correctly. • Generate useful information for diagnosing problems. • Use these tools to: • Continuously document your performance so you know when it changes. • Share your network measurements results.
PerfSONAR Plug • PerfSONAR is a Network Measurement Architecture that is being jointly developed with ESnet, GEANT, Internet2, and a half dozen European NRENS to collect, store and exchange network measurements. • ESnet has developed a proof of concept tool that analyzes the output from traceroute and displays link capacity & utilization graphs for all the links crossed which are available from PerfSonar Measurement Archive servers.
More Info • PerfSONAR • http://monstera.man.poznan.pl/jra1-wiki/index.php/PerfSONAR_About • PerfSONAR Traceroute Visualization • https://performance.es.net/cgi-bin/level0/perfsonar-trace.cgi • OWAMP • http://e2epi.internet2.edu/owamp • IEPM • http://iepmbw.bnl.gov/iepm-bw.bnl.gov/slac_wan_bw_tests.html • http://www-iepm.slac.stanford.edu/bw/ • BWCTL • http://e2epi.internet2.edu/bwctl/