... of disturbance, there's no reason even for multitasking OS to cause latencies more than say in the order of hundreds of milliseconds. Which corresponds to said half of megabyte of buffer, or, say twice so.

If there is no external stimulus - of which the networks are the most unpredictible, that's why I would start with getting rid of that completely - there is no reason for more serious disturbance.

I think it is easier to "purge" Linux than Windows in this respect; and also I believe you can find people capable of doing this, in among those who like to dig in Linux kernel and in substantial Linux packages etc.

On the other hand, I don't think you can get a 100% predictible timing even with DOS, if you allow BIOS to install its interrupt services. And, to overrule that, it's too much of work to be worth. I also don't believe you will get *guaranteed* *sustained* write speed on modern hard-disks either - I might be wrong, of course.

I personally would go for a series of tests with a stripped-down version of hardware I have outlined (basically, just the FTDI chip and a small logic to feed it with any constant data at adjustable speed) on a modern, fast Linux machine. IMHO it will quickly show up, where's the borderline in sustainable speed; and if it's at least 2-3x higher than what you require, IMHO you're done.

JW

PS. What about the option to fail ocassionally, I've asked above?