Wednesday, 19/Thursday, 20 July, 1995 @ 10.00 am
Venue: Heathrow Hilton in London
The members of the Working Group present were as follows: Peter Dzwig - LPAC, Chairman, Matthew Holford - LPAC, Secretary to the meeting, Richard Kaufmann - DEC Ltd, Klaus Wolf - GMD, Malcom Rigg - ICL, Alistair McEwan - LPAC, Denis Caromel - INRIA, Philippe Mussi - INRIA, Jonathon Poole, UCL.
Apologies were received from Heather Liddell of LPAC and from Graham Roberts and Russel Winder of UCL.
The minutes of the previous meeting were distributed beforehand and agreed without amendment.
The meeting agreed to deal with "Matters Arising" on this occasion by email if matters were not covered during the meeting.
A discussion was held during which updates were given by various delegates on the progress of projects which they had been following.
The meeting raised the possibility of developing a taxonomy document with abstracts of the various C++ projects fo the URL of the appropriate papers. RK suggested that of the papers available, perhaps 20-30 should be identified as worth studying. He also suggested that a mailing list be generated of those people reviewing in order that any papers could be distributed. AM volunteered to be coordinator for this exercise.
PED proposed that those reviewing the documents should draft a brief summary of the paper and forward to the Architecture SIG as a whole via email. It was requested that all papers are to be added to the archive. The list of papers for review will then be selected from the archive.
Presentation by KW.
JP said that there was no further developments to report, although a standardisation paper is due to be drafted shortly. He stated that, following three of four drafts, there would be a standard arrived at in a year or so.
2.4 Presentation of Position
Papers
DC made the first presentation on the INRIA position paper on Eiffel ||C++.
RK noted that there might be conflicts between the INRIA system and others (with regard to pre-processing).
JP presented a position paper on UC++. He stated that this position paper represented his own position rather than that of UCL or the UC++ Research Group.
PED delivered the final position paper. He stated that the proposal that he had to offer was directed towards the application writer and that it proposed a framework rather than a language definition. This was agreed to represent the best way forward and the meeting agreed to adopt this approach as the basis of its process.
PED wound up the first day's discussion by suggesting that a shortlist of areas of such a framework/headings for discussion should be identified and further refined on the second day of the meeting, the purpose of which was to provide a breakdown by the end of the meetng of those areas to be addressed. Subsequent to this a draft definition should be prepared and published on the Web soon after the meeting. It was agreed the framework to be worked from would be the LPAC structure incorporating the INRIA model.
The meeting restarted on the second day by outlining the basic structure upon which the EC++ standard document would be based. This represented the outcome of the previous day's delibertions.
- Characteristics of Architectures
- Characteristics of Applications eg system usage patterns
- Language - EC++ should run on any C++ systems unmodified
- Extensions were all to be handled through classes. Use of proxy model attempts to use this to remove underlying hardware assumptions.
- Communications
The proxy mechanism makes no assumptions about communications particulars and/or management scheme implementation. This allows all communications to be abstracted from language semantics.
- Bindings
Bindings should be provided for PVM and/or MPI, for example as a proxy transport mechanism or a message passing system. It was agreed that, although message transport is important no particular protocols will be defined, or suggested, as components of the standard.
- System (Management) Libraries
Naming (system) description
Static scheduling
Load balancing/dynamic allocation strategy
Relationship to existing libraries
- Language libraries
- Tools
Proxy generator tool (required to implement language)
Debuggers/Profilers
- EC++ will not cover these
- Parallel Processing Models
Synchronisation
- Global references from active objects undefined
- Standardised models now
May add in the future
User extensibility
- Semantics of active objects in data members undefined
- Threads
Single or separate thread safe mode
Standard does not address threads explicitly
- Shared data only objects
- Introduction
- Relationship between C++/EC++
DC suggested a standard structure for EC++ based upon 4 levels (see Appendix 1):
EC++ level 0: Proxy mechanism (MOP?) for C++.
EC++ level 1: Library of parallel programming.
EC++ level 2: System management library.
EC++ level 3: Library of binding classes.
AM observed the SIMD was still to be addressed. The meeting discussed the matter and tended to regard it as one which would be addressed through low level libraries or through maths libraries rather than at a language level.
The date of the forthcoming plenary meeting, to be held in Brussels, was agreed as 8 November, 1995 and that this should be published as soon as possible.
It was agreed that there should be at least two technical meetings prior to the plenary meeting with 13/14 September at INRIA in Nice and 17/18 October at DEC in Galway being agreed.
Architecture Menu