François, On Wednesday 02 November 2005 06:31, François Pinard wrote:
[Randall R Schulz]
...
It's true that the memory cache may give the illusion of speed, and that one undoubtedly has to be careful, waiting that all of the floppy has been written for real before removing it.
On the other hand, when a flurry of files are copied to the floppy, those floppy areas holding directories, and information about sector allocation and linking, are being read and rewritten many times, at least once per file written. So, letting the system use the cache brings a real benefit, because all those updates are all done in memory first, and the combined result physically written only once to the floppy.
The system will still use the cache, it's just that writes to the device will not be held indefinitely _only_ in the cache, but rather will be flushed to device as soon as the data is written from the application level. The only real problem (again, bugs notwithstanding) occurs when the application is not writing in sector-aligned quantities. The worst case, of course, is unbuffered, single-byte writing. Line-by-line I/O, while not as bad, will still be far worse than proper sector-aligned writes. To the OP (Pelibali): What program is doing the writing that's so slow? If it's "cp" or "mv" or the default bs= / obs= values of "dd", then things ought to be OK. On the other hand, many other applications will not write in sector-aligned quantities and could trigger the symptom of very slow writing without there being any bug afoot. Randall Schulz