lauren@vortex.UUCP (Lauren Weinstein) (03/30/84)
I think that such statistics should be viewed with quite a few grains of salt. In particular, previous studies on actual telephone circuits (local and long distance) with typical noise problems (static, echo, etc.) showed that the "g" protocol outperformed most other protocols by quite a good margin. In particular, its error correction algorithm outperformed straight CRC, checksums, X.25, (etc.) for actual communications conditions. It was particularly better when it came to the sort of burst errors that make up the majority of errors over telephone lines. By the way, the "g" protocol CRC is a modified CRC-16 and was carefully tested in theory and in actual operations before being made part of the protocol. There are interesting EMPIRICAL statistics as well, including a massive effort made about a year ago when something like 500 meg of data was transferred back and forth across the country via horrible MCI and Sprint circuits (you could hardly hear a voice they were so bad!) via UUCP and then compared byte for byte with the original. The error count was zero. I think that this may be one of those cases where the math and the derived statistics may be misleading compared with the real world. In particular, in my own experiences and the experiences related to me by others over the years, the appearance of corrupt data over a UUCP link is an exceptionally rare event. I have NEVER seen it myself, even during my own massive studies back a ways. So let's not go running around saying that the sky is falling when things are actually working well in the real world! It's the real world operations that actually matter. --Lauren--