Sun Ethernet Behavior

[bob at morningstar.com]

|I have been hearing a large number of complaints about Sun Ethernet
|policies.  Particularly, their effect on a network containing a DEC VAX
|clusters.  

This reoccurs almost as frequently as Craig Shergold and his get-well
cards.  I know this sounds like partisan finger-pointing, but the problem
is in the VAXen, not in the Suns.

|Arguments range from Sun violating the retransmit time to Suns causing all
|types of broadcast storms (ie: ARP requests, NFS packet swamping).

Suns, particularly those with Lance Ethernet chips, are fast enough to
pump out back-to-back Ethernet packets.  This has been known to swamp
other machine's Ethernet controllers if they're unable to keep up, notably
KIs and KLs and older Unibus VAXen.  The Suns are conformant with all the
relevant Ethernet specs, just faster than the older machine designs ever
expected to see on their nets.  The backoff/retransmission rules are
working fine, the VAXen just can't manage the packet rate.

ARP storms might happen if some systems on the network are still using the
old-style (all zeros) broadcast address.  If everything on the wire is
using FF (all ones) broadcasts, then there should be no problem.  The
reason this problem shows up in relation to VAXen is that there wasn't
much besides VAXen running UNIX on Ethernets back when all-zeros
broadcasts were in vogue, and some of those VAXen haven't been
reconfigured since then.  Time marches on...