On Sat, 26 Nov 2011 02:35:17 -0500, you wrote:
3) and from what I remember, they are entirely dynamic DNS based and
thus had to make major changes to the bridge to even exist.
The changes they made work just fine, BTW...
Back then we were entirely dynamic and our network had a full mesh topology.
Now we have both static and dynamic addresses and the network has a full
mesh core plus some "forward legs" speaking only to the nearest core node.
We had and we still have some bandwith and latency issues, especially on
leaf nodes. Because of that, our standard setup has a 90 seconds hello timer
(instead of the default 15 seconds). The same for the multicast LAT timer.
We also had some UDP fragmentation issues with some ISPs, probably due to
their sloppy IP router configurations, or something similar. That is, some
ISPs had (and maybe have) problems transporting UDP payloads that cause the
full UDP packet to exceed the standard Ethernet frame size. We absolutely do
not know why, but we discovered it during long and boring trial-and-error
sessions. Our solution was to reduce the DECnet buffer size from 1498 bytes
down to 1456 or even 1428, so that a full DECnet frame plus the UDP packet
overhead totals to something less than 1500 bytes. It's a tradeoff that pays
very well on the side of DECnet reliability and robustness.
HTH,
G.