- From: <hallam@etna.ai.mit.edu>
- Date: Tue, 11 Jun 96 16:20:04 -0400
- To: "N.G.Smith" <ngs@sesame.hensa.ac.uk>, www-logging@w3.org
- Cc: hallam@etna.ai.mit.edu
I think that there are two separate issues. First we need to establish a base interchange format which will permit exchange of data between different platforms, second we need to determine efficient datastructures for gathering data. A major problem with binary data formats is that the optimum representation is highly dependent on the particular platform. If I have a UNIX system and my file system is limited to flat files I would want a very different setup to that I would want on (say) a VMS system where there is data structure support in the file system. I am also likely to have different requirements for parallel machines. The UNIX architecture is particularly limiting in that the kernel has very limited locking capability. Performing multiple operations as an atomic operation is very expensive. Many of the server architectures (esp. hunt group servers) will have a very hard time of making anything other than an append operation atomic. The problem I see wrt standardizing a binary format is that each platform will have a different optimal format. I don't think we can standardize anything more than the data-structure and leave implementors to work out the optimal solution for their particular case. Phill
Received on Tuesday, 11 June 1996 16:16:12 UTC