> Just some additional data for you. I am seeing this error frequently also. 
> The machine is a MacII using ppp at 28.8, but usually no more than 26k
> actual. I am running X, but have 20 megs of ram, so disk access isn't
> excessive. When data is comming in, I get lots of zs0a:fifo overrun, looking
> at the screen right now, 09:25:42, 09:26:10, 09:26:38. This may sound naive,
> but do these overruns cause loss of data, or do they just slow things down??

The overruns represent loss of data; they don't really "cause" it. As I
understand it, reporting the loss, through syslog, shouldn't cause
fifo errors as interrupts aren't turned off. As the ring errors are posted
in the ring-servicer (the thing which needs to be called to clean the
ring), I don't think the logging should make things worse.

I'm starting to suspect that the only way to sustain full receive rates
(at >=19200 ish) w/o errors (or w/ fewer errors) will be to find
out what (other) part of the kernel is slow, and make it cooperate.
I've read that LocalTalk checks for received bytes on the modem port
during communication. I think it also does something with the disk IO
too. But this big a change won't happen until 1.2. :-(

I'm working on changes to get serial console & echo to work. I'll up the
ring buffer size while I'm at it.

Take care,

Bill