(I can't believe I'm joining in on this discussion, but I'm hoping it'll still die soon) Tim Salo wrote:

...

From: Chris Wilson <cbw@atlantic.net> [...] I was (somewhat) following a dicussion about high speed communications handled over the PCI bus, and there's a company that makes T1/T3-level serial cards for PC's, and from what they were saying, there's no way to reach even a full T3 worth of bandwidth on a PCI bus, so I think trying to push an OC-# connection over it might be pushing it a bit ;)

This doesn't sound right.

As some random data point, we have run ~70 Mbps (ttcp) using ATM as well as Fast Ethernet between 133 MHz Pentiums. I suspect that a faster Pentium would fill an OC-3c ATM link.

The discrepancy is caused by the same limitation that allows the old PC-ISA bus to handle a theoretical 8(?16?)) Mbytes, but start to choke long before a rs232 serial port reaches 115kbps. It's not the processor or the bus, it's the number of interrupts the serial connection is generating. Move away from interrupts, increase the amount of data per interrupt, or whatever, and the limits increase. Doing it at the T level on PCI is just a matter of scale from the 232 level on old PCs with ISA. Discussing raw bandwidth just misses the point. That's my immediate reaction, anyway. Does anyone actually know better how this all works, or are we going to read another 3 days of hearsay?