networks@dartvax.UUCP (Special Projects Group) (10/03/85)
We recently acquired some Sun 50's and hooked them into our Ethernet along with two VAX's running UNIX, one running VMS, and some Sun 120's and a Sun 170. The 50's can communicate perfectly with any of the Suns, but when you attempt to rlogin to the VAX, everything slows to a crawl. The packets/sec drops by an order of magnitude and collisions rise. According to the Sun people on the East Coast the VAX either "talks too slowly for the Sun" or "tells the Sun to slow down and the Sun slows down too much." (These are second hand quotes, unfortunately.) Neither of these makes much sense, but something is definitely happening. Sun did claim that it was a 'feature' and that nothing would be done about it. Seems to me that they should do something about it if they want people to buy 50's. Has anyone else had this problem, and if so, was there anything that could be done? We have someone else calling around Sun trying to get real answers from them as well, but we'd appreciate any advice/help that we can get. Thanks much, in advance. -- David C. Kovar -- Special Projects Group USNET: {linus|decvax|cornell|astrovax}!dartvax!networks ARPA: networks%dartmouth@csnet-relay CSNET: networks@dartmouth US Mail: Kiewit Computation Center Dartmouth College Hanover NH 03755 Phone: (603) 646-3144
gnu@l5.uucp (John Gilmore) (10/08/85)
One problem that earlier Suns had with talking to newer Suns is that the newer Ethernet boards (Sun Ethernet Multibus board, and all VMEbus systems), using VLSI Ethernet chips, can transmit back-to-back packets. The older boards (3Com) only have two receive buffers, and once they fill, until the CPU responds to the interrupt and copies them out, you drop packets. This is aggrevated by each broadcast packet filling one of the two buffers in every 3Com board on the net, making it much more likely that later (broadcast or directed) packets would get dropped. The problem is further aggrevated in the Sun-3 since the data in the packets is generated much faster in the first place. I heard in earlier years that none of the Vax Ethernet boards was very good (though I never worked with one so can't verify it). Once TCP has to go into retransmissions, the thruput goes WAY down, since it's tuned for the claimed 99++% reliability of Ethernet. (I haven't seen a production Ethernet that gets 99++% of the packets through, though...)