170 likes | 312 Views
Grid Network Performance Monitoring for e-Science. Mark Leese - Daresbury Laboratory Thursday 15/5/2003 m.j.leese@dl.ac.uk http://gridmon.dl.ac.uk/~mjl. Contents. Purpose of the work GridMon: what it does & how it does it Progress & examples The future Conclusion Questions?.
E N D
Grid Network PerformanceMonitoring for e-Science Mark Leese - Daresbury Laboratory Thursday 15/5/2003 m.j.leese@dl.ac.uk http://gridmon.dl.ac.uk/~mjl
Contents • Purpose of the work • GridMon: what it does & how it does it • Progress & examples • The future • Conclusion • Questions?
Purpose of the work “…design and deploy an infrastructure for network performance monitoring within the UK e-Science community.” • Fault finding • Performance prediction Key aspects: • Publish results (adaptive) Grid middleware and Grid apps • Visualisation for humans • End-to-end • ability of TCP wrt high b/w networks
Monitor Node 30 mins Tools installed on dedicated & similar node at each centre MESH Monitoring: How(1)? IperfER PingER UDPmon MiperfER bbcp/ftp Publication service Grid middleware Monitoring Architecture www.visualisation
Progress • Intitial toolkit (IperfER, PingER, UDPmon) at 11 of 12 e-Science centres. Other useful UK sites being added. • MiperfER on beta trial at Cambridge, Cardiff, MCC and Newcastle. • Active map also on trial. • Click for AMAZING live demo
The Future (Tasker’s) Trident Longer term: • More sites…but mesh doesn’t scale! • Wishlist features (but perhaps not all 3,000,000) • Investigate other issues: window sizes, QoS… And… Tools www i/f bbcp/ftp GridFTP Web service
TCP Best Practice • Demo TCP best practice to users – “this is what you can achieve, if you ‘tune’ your machine like this...” • Possibles “variables”: • Kernel versions and patches • MTU • Interrupt handling • However: • “...performance from copper-based GigE cards are intimately connected with judicious use and understanding of the corresponding driver(s)...” Gray and Betz • Bus speed, bus width… No new fangled gizmos – using what we already have!
Conclusion • Near national infrastructure • A little basic, but improving • Poised for web i/f into historic data • …. then web i/f tests on demand • What do YOU do next?......... • http://gridmon.dl.ac.uk/
Network Monitoring for e-Science Questions m.j.leese@dl.ac.uk http://gridmon.dl.ac.uk/~mjl ? ? ? ? ?