Partial Minutes for MMN BT URI project for Cambridge, Wed, Nov 5th. 1997 Held at Mo/ller Center, Churchill College Cambridge. Draft of 5/11/97 0/ introductions & apologies Daniel Waddington, unable to make it [Andrew Grace and Rob from BT unable to be here] [David Hutchison is unable to be here til later] Attendees Name, Affiliation, Email Jon Crowcroft UCL jon@cs.ucl.ac.uk 9 James Cowan, UCL, j.cowan@cs.ucl.ac.uk 4 Geoff Tagg, OBU Phyllis Callinan, OBU David Caton Brookes djc@dsrg.brookes.ac.uk 5 Sheil Llloyd Lyons Brookes skll@dsrg.brookes.ac.uk Frank Ball Brookes frank@dsrg.brookes.ac.uk 6 Nat Pryce, IC DOC, np2@doc.ic.ac.uk 3 Morris Sloman, Imperial, mss@doc.ic.ac.uk Chris Edwards, Lancaster, ce@comp.lancs.ac.uk Iain Phillips, Loughborough, I.W.Phillips@loughborough.ac.uk 10 David Parish, Loughborough, D.J.Parish@lboro.ac.uk Omar Basha, Loughborough, O.Basha@lboro.ac.uk Ian Leslie , Cambridge, Ian.Leslie@cl.cam.ac.uk 1 Simon Crosby , Cambridge, Simon.Crosby@cl.cam.ac.uk Sean Rooney, Cambridge Richard Mortier Cambridge Ian Marshall, BT, Ian Marshall Tim Spencer, BT, spencetj@boat.bt.com Simon Carter BT Martin Tatham, BT Labs, martin.tatham@bt-sys.bt.co.uk Andrei Majidian, BT 1/ minutes/actions/agenda bashing Minutes okayed Agenda - added item on Invoicing Actions - a)/Morris had dispensed his by mailing the info on the network management meeting/conference to the mmn-tech list recently b) demos for BT tower event for 7/11/97 cancelled due to change of circumstance (e.g. SJIII contract, or is GTE or is it Worldcom:-) c) New (repeated) Action - please ALL send jon'cs.ucl.ac.uk pointers (URLs) or copies (postscript or html) of their slides for todays presentations!) *** ACTION ALL *** 2/ LEARNET Progress (moved to later so that others from Cambridge and Lancs can be here) Crowcroft resented the status (very nearly have fiber to UCL from BT, have from Essex to BT, and some way to go for Cambridge to BT - next stage is IP connectivity (IP on Raw SDH mainly, although we may deploy IP on ATM on SDH in some places if easier...) Extension to other places (e.g. EC COIAS project, if it happens (see http://www.cs.ucl.ac.uk/staff/jon/coia/index.html) or CAIRN (see http://www.isi.edu/CAIRN/), or London MAN to IC, should be easy. Further sites (e.g. on SJIII) might be possible, but probably by tunnelings over the service net (CWC might not be willing research partners:-) The possibility of machines in BT labs, but outside their firewall, to make URI and LEARNET experiments easier was raised. 3/ work progress reports from each partner Loughborough Iain Phillips presented some updates on the delay measurements on the Superjanet SMDS network - and noted that the performance roles of Lancs and Loughborough have swapped from bad/good to good/bad - he went through some graphs pointing out explanations for loss/delay features (outages, changes in config etc) Omar then went through in detail the architecture for the 2 level system for storing monitoring data, and how the "intermediate" + "primitive" level allows incremental processing of data (intermediate layer stores partially "cooked" data, while the primitive layer stores raw data) there are some nice APIs to this and its a fully OO design, and looks eminently extensible. Crowcroft asked about extracting throughput statistics (e.g. line rate) - for example using 2 packet sizes for pings, and using the pathchar algorithms, to extract the underlying available capacity. In discussing this (see note below) Ian Leslie pointed out that the delay distribution differences for the two packet sizes might well reveal the whole arrival process (pathchar uses a technique for eliminating the arrival process by considering that as a noise process imposed on the underlying constant propagation plus serialisation times for the two packet sizes - Crowcroft noted that on SMDS (or ABR ATM services, or on shared media links) the packets don't see a constant serialisation time due to the access protocol! so we need to consider that too - discussion later between Phillips&Crowcroft decided that though its not obvious WHAT we would learn since the SMDS is largely uncongested, the delays are correlated at least anecdotally with load, so there must be some information about load one could extract, so we should try! ACTION Crowcroft to post info about pathchar *** ACTION *** Lancaster Chris Edwards presented the Lancaster work for Dan Waddington - he discussed both the Video work, and the decisions made for APIs and coding (not to use ASF for example) were justified. Then he discussed the middleware for network control - this work might easily align with similar work (DCAN) at Cambridge. OBU Dave Caton described the change of direction to the OBU work and outlined the heterogeneous multicast work that they are doing - they may continue with the layered video codec as a driver for this and a new PhD student at OBU could start on this - there was a brief discussion about Active Movie versus Video for Windows for integrating CODECS on MS Windows platform based tools. Lancs, IC and UCL preferred the former Dave mentioned that they may use NS for some pieces, but also opnet might be better for low level detailed work - they are looking at receiver driven rate adaption and its interaction with TCP Crowcroft noted that UCL had done some work here in their hipparch project - see recently accepted INFOCOM paper at ftp://cs.ucl.ac.uk/darpa/infocom98.ps Crowcroft also talked about adaptive protocols and the work UCL is doing on this in the area of pricing for TCP performance by extending the ideas from Frank Kelly's Rate proportional priced sources, (see http://www.statslab.cam.ac.uk/~richard/PRICE/pricing-internet/ especially http://www.statslab.cam.ac.uk/~frank/rate.html and of Sally Floyd, Matt Matthis& Tunis Ott in ACM CCR earlier this year (linked above) Crowcroft and a post doc visitor at UCL, Philippe Oechslin, will give a talk about this at BT soon (hope we can post a paper on this before long). Cambridge Ian Leslie presented the Cambridge related work (e.g. DCAN, Networks on Demand, LEARNET etc) and explained that there would be a hiatus in the Cambridge work for the next few months, butt that this should not create any problems for the other partners (he outlined the way this would be ok), and that they would restart after the inter-regnum in January. IC Nat Pryce presented the Darwin threading work, and the IPC work called MIDAS< based on the Corba Idl but with a more subtle and expressive interface and behaviour model UCL James cowan presented the Internet protocol stack management work - we have a rate controller installed at UCL and BT Labs, and there's working control interfaces in CORBA, Java RMI, and a DCOM one in progress. In the BT testbed, it is the experience that the Java Gui is the biggest performance hit, and that the stack manager overhead is minimal! 4/ Patent Progress We talked about the value of patents, both to BT and to partners. Crowcroft expressed concern about the slowness of exploitation since we tended realistically to rely on BT to get some of their system providers to exploit things, with the current way of working. Ian Marshall responded that once a patent is through, we (the universities) can work with provider companies and BT will be happy with that... 5/ Integration Plan We need to direct some more effort now into this, since we would like to have a reasonable show of this (probably on LEARNET...) - we should discuss the work at lancaster, ucl and imperial particularly in this context - i'd like to see what we can write down now for a strawman demo (to be monitored by loughborough...) Crowcroft and Marshall proposed that a model for a two level demo, with loose coupling between the levels seemed most appropriate IC, Lancs and UCL have end system work (Qos, policy, stack and access link management) Cambridge, OBU and Loughborough operate mainly within the network So we can build a demo with video and audio services and conventional TCP (e.g. web) based applications, whose policies are managed by IC and whose stacks are managed by UCL, given some feedback from the Loughborough network system. Dave Parrish asked about how the overall network is to be controlled. Crowcroft proposed that a nice way to glue this together into a coherent whole is to picture pricing (markets) as a distributed controller, and use the LEARNET work there to control networks on demand, through the UCL and Cambridge work, then monitor within these at a finer timescale, so that applications can tell what type of performance they are getting, and, if within policy, adapt. Policies could also be applied to how people "spend" money to ask for more capacity in a longer timeframe... The usual question was asked about standards for interfacing (CORBA Idl/IIOP, versus Java RMI versus raw sockets etc) - *** ACTION ALL *** Keep up ongoing discussion of details for integrated demo....to be sometime in early-mid 99. Two things to be resolved soon 1/ interfaces (pairwise) between pieces - first one maybe to tackle is getting feedback from Loughborough system to UCL and IC and Lancs! maybe a JAVA RMI and an IIOP interface could be built by folks from UCL and IC and LANCS helping Iain and Omar? 2/ the timescales of monitoring (e.g. do we use them for building a network on demand with the Cambridge work, as well las for feedback, or just the latter? Crowcroft and Crosby agreed just the latter. 3/ The pricing model needs to be imported from the LEARNET project - e.g. how is subscription pricing subjected to CAC and how does this influence the network on demand building (e.g. virtual link placement and dimensioning) algorithms? This is not a key piece of the demo though! *** ACTION Ian Marshall ** * to produce two draft documents outlining the network and end system demo models.....with input from partners! 6/ T&D next meet Wednesday, 4th March, loughborough (to be confirmed) 7/ AOB Invoicing! 3 Universities are ok IC hasn't claimed for last period UCL hasnt either and is also using the previous contract number OBU appears to be using random contract numbers to claim. *** ACTION *** Please can UCL and OBU use the right contract numbers, and the right office (Manchester) and so forth.