QoS Trials on SuperJANET

in the Context of HICID and other BT/JISC project

Jon Crowcroft and Peter Kirstein

November 14, 1997

 

 

This proposal is to JISC to permit network level experiments on the SuperJANET network. The purpose is to permit UCL to carry out some experiments in using next generation IP traffic service facilities in Ciscos (and other routers) on some four Sundays over the next three months. We do not intend to bring down the network as part of these experiments. Nevertheless, during the experiments certain parts of the network must be considered "at risk"; because we are requesting that this occur only at specific planned times on a Sunday, we believe that the possible inconvenience to other users of SuperJANET can be minimised.

 

Currently we run two application sets at two or three sites over the normal UK Mbone; one is a conferncing set Vic/Rat/Wb/NTE, and the other a Video-on-Demand application running MMCR or PRECEPT software. For these experiments we would run the same experiments - but would turn on Random Early Drop (RED), or weighted RED, with precedence queuing on a specific path(s). Two of the paths we expect to use are UCL<-> Bristol and UCL <-> Essex both in the standard Ciscos and with other routers. These routes are chosen because they are participating in the BT/JISC funded HIGHVIEW and JAVIC projects; this work would be carried out under the HICID project which is funded under the same programme.

 

  1. CISCO Experiments

 

The aim of the activity would be a co-ordinated measurement experiment, using only the Standard Ciscos, to investigate the following:

  1. The improvement (i.e.. lower latency and lower loss) ) for precedence queued traffic by monitoring the playout buffers and subjective testing the interactivity;

  2. The decrease in available capacity for non-precedence traffic;

  3. The lack of overall harm to non-precedence traffic

iv) The improvements for ALL traffic when RED is enabled.

 

  1. CAIRN Router Experiments

    2.  

    Once we have done the above, we would expect to install CAIRN routers outside the standard SuperJANET and repeat the experiments. Here we expect to use the same three sites (plus BT and Cambridge) for the purpose of investigating:

    i) Better multicast routing;

    ii) Traffic monitoring ad tagging

    iii) Experimentation with new replacements for RSVP (and CBQ)

    iv) Experiments on pricing (for LEARNET, HIGHVIEW etc)

     

  2. IP/ATM Integration

 

As a final stage, we will experiment with the Cambridge/UCL work on IP/ATM integration. This activity depends on using the PC platform for IP to ATM proxy signalling (e.g. when IP traffic arrives that matches some precedence or other class, we signal an ATM layer from the IP router (rather than from the source host)

 

This will be partly based on DIMMA and DCAN work from Cambridge Computer Lab, plus work by Paul White here, and new PhD students, Wayne and Dirk here.

 

We do not expect the third part to work for at least a few months from now. It requires later versions of LEARNET, following earlier experiments with Cisco and Fore kit. Note that the SuperJANET net between the two of the sites and London (Bristol and Essex) is not IP/ATM but IP/SMDS.