End to end performance networkshop44

End to end performance networkshop44

• 1.
  End-to-End Performance Share andExplore Tim Chown
• 2.
  Introduction » The aimof this ‘birds of a feather’ (BoF) session is to facilitate open discussion on a variety of end-to-end network performance topics › Follows up on the presentation session earlier today › Identify common issues, and discuss potential solutions, or – perhaps – further work / priorities for Jisc » NB. If you are tweeting from this BoF, please respect ‘Chatham House’ rules › i.e. relay comments, but not who made them 22/03/2016 Title of presentation (Insert > Header & Footer > Slide > Footer > Apply to all)
• 3.
  Topics » Topics thathave been suggested include: › Use of local site network engineering › Appropriate tools for data transfer › Network performance monitoring › Knowledge sharing on system tuning › Interacting with users, PERTs, … › +....? » General assumption that the network core is capable › Problems *tend* to be at the ‘ends’ 22/03/2016 Title of presentation (Insert > Header & Footer > Slide > Footer > Apply to all)
• 4.
  Local site networkengineering » There has been some suggestion recently that sites should seek to implement a ‘Science DMZ’ approach › A ‘controlled firewall bypass’ – In principle it can be beneficial, but many issues involved » Use of a site’s ‘resilient’ link for bulk data » Internal use ofVLANs/MPLS for traffic segregation » How extensive are such methods already? › Useful to understand this › (see informal questionnaire…) 22/03/2016 Title of presentation (Insert > Header & Footer > Slide > Footer > Apply to all)
• 5.
  Appropriate tools fordata transfer » We can learn a lot from the tools being used by HPC communities, e.g. GridPP’s FileTransfer Service (FTS) › Can/should we create something ‘lightweight’ that can be distributed to other communities? – Would require some committed effort… » Perception exists that too often researchers try scp or similar ‘simpler’ tools, get poor results, and give up › Part of issue here is knowing where to turn for help › And expectation management » What tools are people using? › Is there interest to do better here? 22/03/2016 Title of presentation (Insert > Header & Footer > Slide > Footer > Apply to all)
• 6.
  Network performance monitoring »When presented with reports of poor performance, it’s desirable to get some level of telemetry on the problem › Ideally as close to the application as possible » Various possibilities, e.g.: › Full scale perfSONAR › ‘Small node’ perfSONAR (on LIVA X type system) › Roll-your-own iperf, etc › RIPE Atlas probes › … » What are people using now? 22/03/2016 Title of presentation (Insert > Header & Footer > Slide > Footer > Apply to all)
• 7.
  Knowledge sharing onsystem tuning » If we can eliminate as far as possible performance issues within the network elements, then we can focus on end system tuning › Can you get the desired performance locally between two systems? » A good amount of guidance out there today › See http://fasterdata.es.net/ in particular › TCP parameter tuning, etc » Various scales of problem › Depends on the nature of the end system » Are we involved in helping our own users with such tuning? › If so, how well do we do? 22/03/2016 Title of presentation (Insert > Header & Footer > Slide > Footer > Apply to all)
• 8.
  Interacting with users,PERTs » One of the challenges in delivering end-to-end performance is getting the user communities, local campus computing services, and the Janet NOC working together effectively › Are users reporting issues? › Is the local computing service aware of sources of help? » Some NRENs have an explicit PERT function › eduPERT database: http://services.geant.net/edupert/Pages/Home.aspx » At the moment, the Janet NOC is implicitly the Janet PERT › But we are also building expertise through the Janet e2e initiative » Are we doing as well as we can here? 22/03/2016 Title of presentation (Insert > Header & Footer > Slide > Footer > Apply to all)
• 9.
  Any other topics? »Anything else to discuss? » Jisc is planning a workshop on local site network engineering, probably in London in June › Presentations and discussion on current practices, and how we might do appropriate network engineering to support future data intensive research › How does the ‘Science DMZ’ approach fit in this picture? » Possible ESnet OIN workshop in Cambridge this summer › See http://oinworkshop.com/ for nature of content » Thank you for coming  22/03/2016 Title of presentation (Insert > Header & Footer > Slide > Footer > Apply to all)
• 10.
  jisc.ac.uk 22/03/2016 Title ofpresentation (Insert > Header & Footer > Slide > Footer > Apply to all) Tim Chown Tim.chown@jisc.ac.uk