Re:Propagation Delay and Its Relationship to Maximum Cable L...
by jcr on Thursday April 25, 2013 @07:58PM
(#5610728)
> Exactly. The switches buffer packets.
In most cases, you're right. The vast majority of switches use a sore-and-forward design (buffering), but when you get into latency-critical applications, another type of switch design is used, typically called a cross-bar switch.
http://www.researchgate.net/publication/220055485_A_low-late...
http://www.google.com/patents/US5339396
I think opinions vary on whether or not crossbar designs "buffer" but it mostly depends on how you define buffering. None the less, the low-latency crossbar designs try to eliminate the latency caused by buffering in the older but more common store-and-forward designs.
Re:Propagation Delay and Its Relationship to Maximum Cable L...
by ajross on Friday April 26, 2013 @12:05PM
(#5613647)
This is a valid point and I should have been more precise. But AFAICT all those interconnects are single duplex point-to-point links, and thus not subject to the CSMA issues of asymmetric collision detection detailed in the linked article.
And my experience is that you can actually get well beyond the official lengths for most signal types even with comparatively cheap cables as long as you're willing to test and throw out the odd cable or connector. Standards are written with conservatism in mind.