Rules for Penalty calculation

There are two sorts of transformation: (a) conversion from one format into another, for reasons of presentation, and whereby information generally is lost, and (b) encoding, for reasons of speed (compaction), security (encryption), or transmission, and whereby the information and format remain untouched.

There are two questions to consider when deciding on different possible transfer formats between servers and clients: Information degradation and elapsed time.

Degradation

When information is converted from one format to another, it may be degraded. For example, when a postscript file is rendered into bitmap, it loses its potentially infinite resolution; when a TeX file is rendered into pure ASCII, it loses its structure and formatting.

This degradation is difficult to guess from simply the file type. and for a given file it is quite subjective. Any attempt to estimate a penalty will therefore be very aproximate, and only useful for distinguishing widely differing cases. A suitable unit would be the proportion, between 0 and 1, of the information which is not lost. Let's call it the degradation coefficient. One would hope that these coefficiemnts are multiplicative, that is that the process of converting a document into one format with degradation coeficient c1 and then further converting the result of that with coeficient c2 would in all be a process with coeffcient c1*c2. This is not, in fact, necessarily the case in practice but is a reasonable guess when we know no better.

Elapsed time

The elapsed time is another penalty of conversion. As an aproximation one might assume this to be linear in the size of the file. It is not easy to say whether the constant part or the size-proportional part is going to be the most important. The server, of course, knows the size of the file. It can in fact as a result of experience make improving guesses as to the conversion time. The conversion time will be a function also of local load. For particular files, it may be affected by the caching of final or intermediate steps in a conversion process. Given a model in which the server makes the decision on the basis of information supplied by the client, this information could include, for each type, both the constant part (seconds) and the size-related part (seconds per byte).

_________________________________________________________________

Tim BL, RC