UCL Activities on the CAIRN Network to September 1998
Peter T. Kirstein and Jon Crowcroft, 07 July 1998
In this report we briefly describe the short-term development and deployment experiments in the context of UCL's participation in the CAIRN project.
The current situation is that we have a fat Pipe configured as shown below:
UCL GSFC
CBQ
ISIE
Figure 1 The current Fat Pipe Situation
Currently the Fat pipe consists of two ATM VCs between UCL and GSFC, and one between GSFC and ISIE. There are ATM switches at UCL, GSFC and AISIE. One VC from the UCL CISCO is passed through to a CISCO at GSFC, the other from the CAIRN router at UCL terminates on a CAIRN router at ISIE. At GSFC, the CISCO router is connected on to NREN; At ISIE, the CAIRN router is attached to CAIRN. At UCL, the CISCO is attached to SuperJANET.
We are currently on the process of deploying Class Based Queuing (CBQ) on the fat pipe. There are two frameworks for QoS available; one is the widely available CBQ implementation in the ALTQ distribution from Sony; the other is that available from. Hui Zhang of CMU. There are some problems with both. The ALTQ version does not yet support IPv6, and does not work with the CAIRN kernel; there is some doubt whether it supports the ATM card used, but it probably does. It is not yet clear when it will support the CAIRN kernel. The CMU release does work the CAIRN kernel, but is not quite as complete on QoS algorithms. In both versions, there are implemented several queuing and scheduling disciplines by changes at the interface level of the BSD networking code and the appropriate drivers.
We have worked so far with ALTQ, and we have tried to integrate ALTQ support in the CAIRN version of the ATM driver. Our intention is to enhance the CAIRN testbed with a RED gateway support and Explicit Congestion Notification (ECN) along with CBQ, enabling a wider range of network experiments to be conducted.
Until recently there has been no support for alternate queuing in general and in the CAIRN kernels in particular. So it has to be integrated with the currently supported features of the ATM driver provided by ISIE (IPv6, BPF, per PVC interfaces) which makes the CAIRN stack a special case. We expect in the near future to do some IPv4 experiments with the ALTQ/IPv4 stack, and some with the CMU/IPv6 stack to see which is the more flexible and useful.
One short-term plan is a three-way experiment between UCL, ISIE and CMU using the CAIRN VC to UCL. Another is to experiment with CBQ and gated between UCL and NASA GSFC and to demonstrate that class-based queuing can be deployed for effective bandwidth sharing between agencies (CAIRN, NASA traffic) and protocols, destinations (unicast, multicast). For this purpose we are going to use the UCL PVC and then move onto merging NASA and CAIRN traffic using single PVC using link-sharing provided by CBQ.
For the ALT-Q experiments, we have started liaison with NASA GSFC (George Uhl) and we will make the UCL PVC terminate at GSFC (instead of ISIE) for QoS testing between the two labs. In this case, UCL will either temporarily lose connectivity with the rest of CAIRN or have temporary connectivity through the GSFC CAIRN router. The second option may be more difficult to manage administratively. In either case we expect to see the experiments going on for a few weeks. ALTQ only, an experimental hybrid CAIRN with ALTQ kernels, and the CMU kernel will be tested.
For the work with CMU, they provide a different form of QoS already built in the CAIRN kernels based on packet classification at the IP level (as opposed to the ALTQ approach). It performs link sharing and schedules IP packets according to the Hierarchical Fair Service Curve scheduler (as presented at the SIGCOMM'97 paper). The initial experiments between CMU and ISI-E will be extended to UCL during August. A comparison between the two link-sharing mechanisms will be interesting.
We have committed to present the capability for link-sharing and the results of all these experiments in a technical report by the end of the first project extension.
The CAIRN experimenters have expressed interest in using the latest UCL tools. We expect to release a new version of RAT, with the capability of video synchronisation, and a high quality audio Codec. Like all versions of RAT, this includes packet redundancy of various types, and support for Windows'95/NT. We have also modified VIC to include the Conference Bus compatible with the RAT one, so that audio/video synchronisation should now be straightforward. We have also upgraded the versions of NTE as a text editor, and wbd as whiteboard support for PCs. We expect to release all the above by the end of July, and carry out experiments using them in August over CAIRN.
We plan security activities both in the short and long term over CAIRN. However the longer term ones depend partially on the second extension of our current DARPA contract. In the short term, we have ensured that the various tools (RAT, VIC, NTE and possibly wb/wbd) can be encrypted. The encryption can be both manual, by entering a Pass-Phrase, and automatically from the Session Announcement tool SDR. The SDR has also been extended to allow authenticated and/or private announcements - the latter both with symmetric encryption as before, and with asymmetric encryption. The latter requires the distribution of Group key pairs by PGP or S-MIME. We expect to deploy this complete system by the end of September, for experimentation by the CAIRN community. As part of this work, we also plan to provide a depository of certificates for any interested party in the CAIRN community. This cache will probably be held in a Secure DNS.