Work plan and Project log

The work plan is taken from the approved EPSRC case for support and may not bear any relation to reality. The purpose of putting information here is so that we can write a sensible report at the end of the grant period.

Given that the project actually started spending money (in Bath) on October 15, 1994, I have divided the work plan sections up into quarters, starting with 4th quarter 1994.

1994: 4Q94

1995: 1Q95, 2Q95, 3Q95, 4Q95

1996: 1Q96, 2Q96, 3Q96, 4Q96

1Q97

plan:

algebra applications; statistical computing; AI applications; distribution;

actual:

Meeting in Southampton on 05/01/97. Present: Julian, Russel, Dave, Hugh Glaser, Danius, Stuart McLean, Luc.

Reports:

Danius is joining us as a research assistant on the Jisc (sp?) project, involving Manchester, Chester, and Southampton. The goal of the project is to study/design/evaluate a cluster computing solution for PCs running Linux, Windows95, or Windows/NT.

Andreas is in Pisa working on compiling youtoo to the Java Virtual Machine (JVM) and/or on a port of youtoo to Windows/NT. Andreas will be back on the 20 March. The new release of youtoo was delayed due to a crash of the Denton disk. Everything is fixed now!

During his sabbatical, Dave was thinking to do a feasability analysis of the port to Windows NT. Julian mentioned that a guy in Leuven volunteerd to do the port himself. We will review the situation in late March or early April and then take a decision.

Luc enquired about the availability of Noostoo. Julian said that the system is available, as developped by Francisco, but probably needs a documentation before being released.

Dave's Dagsod project concerns genetic programming. Two applications are involved: one developped in C at Southampton, the other written in Common Lisp, the "standard" benchmark for genetic programming by Kosen. Dave suggested that an examplar for the project could be a distributed Lisp version.

We also reviewed the different examplars that Denton should produce. Bath have developped several AI examplars: OOPS5, NOOS, and the fish market; they are also suggesting a computer algebra one. Southampton have developped the statistics examplar, and are suggesting a Multimedia application or a Genetic progamming examplar.

Julian and Russel described their ATM configuration. Russel evaluated the performance of their network. 1) Local ethernet: 1.5 Mbits/s (old sun 4) 6Mbits/s (Indy). 2) Through a gateway: 1.5-2 Mbits/s. 3) TCP on top of ATM: 35-53 Mbits/s. 4) To local host: 50 Mbits/s. 5) TCP on top of LANE on top of ATM: 8-18 Mbits/s.

Julian described a final year project related to the Denton Project. The idea to is to use an http cache as a medium to communicate between distributed youtoo environments. In the http protocol, Post is a non-cache command, which is passed to the http server behind the cache. The goal of the project is to define a cached version of the post command. Information could be retrieved by a get. This functionality would be reified in youtoo as a stream of htpp commands.

The natural question that follows is to decide which programming model would be suitable for this communication mechanism. Is this a Linda pool?

There is also another interest for http: Luc and Dave would like to define a communication module based on http for Nexus. As a result, Nexeme could be used to communicate over http, across firewalls. Dave is planning to do that implementation in March during his sabbatical.

Stuart McLean presented the GraphIcslas approach for distribution. It consists of a pool of tasks that can be stolen by "idle" processors.

Denton III:

We concluded the meeting by a discussion about a possible Denton III. This a list of possible topics for a research proposal.

Artificial Life, Computational Biogology.

MATHFIT: As part of its continuing support for the health of both IT and Computer Science and Mathematics, the EPSRC wishes to foster and improve the links between these two disciplines through the Mathfit initiative. Mathfit is jointly sponsored by the EPSRC and the London Mathematical Society (LMS), and will run for 3 years. It will encourage the submission to EPSRC of proposals for cross-disciplinary research and Visiting Fellowships, and fund Summer Schools and workshops which address priority areas.

Business Process Modelling

Games (cf. agents for playing diplomacy)

Business over the Internet (ESI: Electronic Share Indicator: www.esi.ac.uk)

Data Mining over the Internet

Sound

Interpreters for Distribution. This project would build upon the technology developped in Denton III. Several aspects could be studied: Instrumenting Distributed Interpreters (cf paper about Interpreters at Asplos); Optimising/Reengineering Byte code (in particular JVM); Security (cf POPL97 paper); Information leaks (cf. trust analysis by Orbaek & Palsberg (Aarhus)); Mobile Code (code a patte!);

Actions:

Dave

to implement a communication module over http for Nexus.

Luc

to enquire about the status of youtoo.

Julian

to enquire about Biological Science at Bath.

All

Review the status of the Windows/NT port.

Next Meeting:

Pencilled date is the 5th of March.

Actual:

Meeting in Bath on March 5th, 97. Present: Julian, Russel, Rob, Dave, Luc, John Sharp

Reports:

John described his work on http caches. He changed the command put (or was it post) so that an object could be stored into an http cache and could be retrieved using the usual get command on a URL. From that followed a discussion on coherence, causality, access to information, synchronisations, diffusion trees.

Luc presented the garbage collector that he implemented in NeXeme.

Luc reviewed his experience with youtoo 0.91, and Julian gave Andreas' answers.

We concluded the meeting by a discussion on Denton III. Julian presented PCC, proof carrying code, presented at POPL97. Julian suggested to write a proposal centered around the simulation of "Electronic Market places". This idea was generalised by Dave who suggested "virtual worlds".

Next Meeting:

Pencilled date is the 2nd of April. (I checked: this is a work day, so I assume that the University is open!)

2Q97

plan:

algebra applications; statistical computing; AI applications; distribution; evaluation;

Actual:

Meeting in Southampton on April 23rd, 97. Present: Julian, Andreas, Dave, Luc, Danius.

Reports:

Julian described four undergraduate projects that he supervised this year. 1) The goal of the first project was to build a graphical user interface for an SQL database. The interface was built using TCL/TK and youtoo linked with the mSQL API, which is talking to the mSQL daemon via sockets. 2) The second project was to build a data visualiser for youtoo. The use of the metaobject protocol is convenient to write generic walkers. Two PD packages were studied: envdraw based on Stk and Davinci based on TCL/TK. The former seemed to display graphs in a nicer way than the latter. The conclusion is: it remains difficult to display graphs! 3) The third project consisted in building a distributed MUD using youtoo and MPI. Interaction is via a Netscape session. 4) The fourth project is the integration of http caches with youtoo. Three operations are available: put, get, and in. URL's are composed of the DNS name, the process id, and a user tag. Information can be retrieved from the cache by giving the key (wildcards are allowed).

Andreas described the work he did in Pisa. He tried to offer a better integration between youtoo and java. He presented three different approaches: 1) link Java VM and youtoo VM and offer cross platform calls 2) write a component and integrate it using java beans 3) use RMI. He adopted the third solution and wrote a serialiser/deserialiser for youtoo that was compatible with java. In addition, from youtoo, it is possible to call a java method.

Andreas and Julian described the new hierarchy of streams in Eulisp.

Luc's update about quantum was delayed after the Vim meeting.

Again, we concluded the meeting by a discussion on Denton III. We started a discussion about Virtual World. Dave mentioned the work by Tom Rodden (http://www.comp.lancs.ac.uk/computing/staff/tam.html, http://www.comp.lancs.ac.uk/computing/research/cseg/projects/virtuosi.html) and Steve Benford (http://www.crg.cs.nott.ac.uk/~sdb/, http://www.crg.cs.nott.ac.uk/Virtuosi/

Next Meeting:

Nothing decided!

Action:

Dave to install gcc and bash on an NT machine.

If someone from Bath goes to ICFP/PEPM, please buy a copy of ICFP proceedings for Soton.

the talk about auctions "Efficient Resource Allocation Through the Use of Auctions", http://www.cs.dartmouth.edu/~jonathan/talk/contents.html

Actual:

Meeting in Bath on May 28th, 97. Present: Julian, Russel, Andreas, Rob, Luc, Danius.

Reports:

We reviewed the situation of the bignum library. It is just a matter to package it and include it in the distribution.

We reviewed the situation of two bugs of youtoo, related to the foreign interface and a "display" error. Andreas has given more information about the foreign interface, and Luc has to find out whether it works now. The "display" error remains to be fixed.

Andreas explained how functions are serialised in youtoo. He uses a "fine grain" approach, where the code of the function only is serialised (and not the code of the class or module).

Luc described the current implementation of Quantum, and the new results obtained with Christian. More information is available at http://www.ecs.soton.ac.uk/~lavm/Quantum.

We searched the web for more information about virtual worlds. We try to find information about infrastructures and languages (i.e. abstractions) for virtual worlds. We distinguish virtual worlds from virtual reality: virtual reality implies the immersion of the user as opposed to a virtual world, which is populated by agents. Somehow, we would like to find the building blocks that specify how agents can interact in such a virtual world. Also, we don't want to do cscw.

Next Meeting:

Jul 9th is pencilled

Action:

3Q97

plan:

algebra applications; statistical computing; AI applications; distribution; evaluation;

Actual:

Meeting in Southampton on Jul 9th, 97. Present: Julian, Russel, Andreas, Dave, Luc, Danius.

Reports:

Andreas has released youtoo version 0.93. The distribution includes Boehm's collector and OOPS5. The compiler was improved; in particular, arithmetic operations were optimised for the case of binary integer operations.

Dave reported on his visit at MIT. He described the Amorphous Computing project led by Sussman and Abelson. An amorphous computer is typically based on 10000 processor units communicating by unreliable local broadcast. Communication operations available to the programmer are local broadcast and receive. At MIT, they have implememted a simulator for such a computer and they are also building some hardware for it. Dave experimented with this new style of computer. He programmed a global broadcast mechanism bound by a hop counter, limitating the broadcast expansion, and he implemented a reliable stream mechanism between two processors. The latter experiment was rather unsuccessful as the amorphous computer is not able to broadcast in a given direction.

Julian described his work on the CEK machine and A-Normal Form. Fist, he presented the CEK machine for an extended lambda-calculus. Then, he described the specialised variant of the CEK-machine for A-NF. Then, he showed that the transformation to A-NF could be encoded as explicit transition rules of the abstract machine: the CEK/WC machine converts and evaluates expressions on-the-fly. Finally, using a staging transformation, he was able to derive a set of rules that perform the A-NF transform and a set of rules that perform the evaluation reductions. He compared the efficiency of the different machines using the number of transitions and activation frames for various programs. His presentation was followed by a discussion on future work.

Next Meeting:

Aug 6th.

Action:

none

Actual:

Meeting in Bath on Aug 6th, 97. Present: Julian, Russel, Andreas, Dave, Luc, Claudia.

Reports:

Julian and Russel have modelled the Fish market in the pi calculus. They added two new operators to the core of Milner's calculus. The broadcast operator allows them to send a value to several channels, while the "gather" operator is able to receive a value from several channels. The buyer, the remote control and the auctioneer are each represented by processes. The presentation was followed by a discussion of tools to run/animate Pi-calculus programs. Amonst others, we mentioned the join-calculus, Peter Henderson's hoc, Pict, Facile, CCS workbench, mobile calculi home page. Luc has also written a cps evaluator with synchronous send and receive, and asynchronous broadcast.

Andreas has been comparing the performance of youtoo with other Lisp/Scheme implementations. He described a new set of benchmarks. The results are available from Here.

Next Meeting:

Sep 3rd (During the PLILP conference)

Action:

none

Actual:

Meeting in Southampton on Sept 3rd, 97. Present: Julian, Russel, Andreas, Dave, Luc, Danius, Manuel Serrano.

Reports:

Manuel Serrano presented his PLILP paper about inlining.

Julian described the results he obtained with Russel on the modelisation of the fish-market in the Pi-calculus. This was followed by a discussion on the limitations of the pi-calculus. A major problem encountered was scalability/modularity as huge lists of parameters had to be passed in argument. Another problem was to be able to define suitable abstractions, and how to guarantee that they compose reliably.

Luc sketched the implementation of the naming game he did with Claudia. As the implementation is distributed, the experience differs from the one done by Luc Steels, because games are not totally ordered. A very brief discussion followed. It emerged that it was not too important if the experience was not exactly the same as we do not extract precise values from it, but observe a general evolution. Danius suggested to see how much the experience depend on the averaging process that is currently used.

We attended Guy Cousineau's invited talk at PLILP97 (functional programming and geometry).

Next Meeting:

??

Action:

none

4Q97

plan:

algebra applications; statistical computing; AI applications; distribution; evaluation;

actual: