stacy@sobeco.com (s.millions) (01/09/91)
Can some one explain to me why mips runs some of thier machines
with trailers off (Rx3230 m120 rx2030 and some m2000's) and some
with trailers on (RC62* and the rest of the m2000's).
I expect it is a performance issue, but am wonder how big the
difference will be. I am having to turn off trailers on our
RC6280 so that it can talk to our DEC 20's.
-stacy
--
"Sorry I had to plug you mister duck, but I'm a sportsman." stacy@sobeco.com
- Almyer Fudd uunet!sobeco!stacywje@redwood.mips.com (William J. Earl) (01/11/91)
In article <1991Jan9.144544.22626@sobeco.com>, stacy@sobeco (s.millions) writes: > Can some one explain to me why mips runs some of thier machines > with trailers off (Rx3230 m120 rx2030 and some m2000's) and some > with trailers on (RC62* and the rest of the m2000's). > > I expect it is a performance issue, but am wonder how big the > difference will be. I am having to turn off trailers on our > RC6280 so that it can talk to our DEC 20's. RISC/os turns off trailers by default on some systems (generally those with an integrated Ethernet interface), because hardware considerations make trailers less efficient. On other interfaces (generally the VME-based interfaces), there is little or no difference in performance, so we allow trailers, in case some other host runs better with trailers. That is, RISC/os always accepts trailers on input, but advertises trailers only when there is no performance penalty. There is, in any case, no performance penalty to disabling trailers between RISC/os systems, so you can freely turn them off on your RC6280. -- William J. Earl wje@mips.com MIPS Computer Systems 408-524-8172 930 Arques Avenue, M/S 1-03 FAX 408-524-8401 Sunnyvale, CA 94086-3650
datri@convex.com (Anthony A. Datri) (01/11/91)
I've seen machines running with trailers on screw up other machines on the net. I recommend always turning them off. -- -- 336996351123355369963511235117*47*8*47*8*7471
bin@primate.wisc.edu (Brain in Neutral) (01/11/91)
From article <1991Jan10.203530.7133@convex.com>, by datri@convex.com (Anthony A. Datri): > I've seen machines running with trailers on screw up other machines on > the net. I recommend always turning them off. Exactly. Mips should turn them OFF on every machine it ships. -- Paul DuBois dubois@primate.wisc.edu
erik@srava.sra.co.jp (Erik M. van der Poel) (01/12/91)
> I am having to turn off trailers on our > RC6280 so that it can talk to our DEC 20's. Do you have to turn off trailers on your RC6280 so that it *can* talk to your DEC 20's, or so that it talks *faster* to them? - -- Erik M. van der Poel erik@sra.co.jp Software Research Associates, Inc., Tokyo, Japan TEL +81-3-3234-2692
stacy@sobeco.com (s.millions) (01/14/91)
In <4922@srava.sra.co.jp> erik@srava.sra.co.jp (Erik M. van der Poel) writes: >> I am having to turn off trailers on our >> RC6280 so that it can talk to our DEC 20's. >Do you have to turn off trailers on your RC6280 so that it *can* talk >to your DEC 20's, or so that it talks *faster* to them? It was done in an attempt to get them to talk, just talk. As it turns out the trouble had nothing to do trailers. The MIPS device driver for the eagle ethernet controller tries to do block mode transfers when ever possible. If mbuf length is not a multiple of 32 bits, they will ship out 1-3 bytes of random garbage at the end of the packet. The dec 20s take the length of the ethernet packet - length of headers and calculates its checksum on that, including the random chars at the end which are not part of the data. As a result the dec 20 throughs out the packet. This was a nuisance to track down. As luck would have it my login name and password made mbufs that were multiples of 32 bits in length as were the files I transfered as a test, so it would work for me, but not for others. I have fixed this by brain damaging the device driver to not use block mode on out bound packets. Will try a more elegant fix when time allows. -stacy -- "Sorry I had to plug you mister duck, but I'm a sportsman." stacy@sobeco.com - Almyer Fudd uunet!sobeco!stacy
rogerk@mips.COM (Roger B.A. Klorese) (01/14/91)
In article <1991Jan13.214030.27744@sobeco.com> stacy@sobeco.com (s.millions) writes: >I have fixed this by brain damaging the device driver to not use block >mode on out bound packets. Will try a more elegant fix when time allows. RISC/os 4.52 includes a more elegant fix. For RC62x0 systems, multiples of two will be used; the DMA system allows block mode then for all but the final partial word, and the last junk byte is handled correctly on other systems. For non-RC62x0 systems, going to a multiple of two will disable block mode and will slow Ethernet transfers, so we will allow a kopt option to control whether or not roundup is done. If you have the extra-bytes problem on a non-RC62x0 system, you may disable roundup, at the cost of Ethernet speed. -- ROGER B.A. KLORESE MIPS Computer Systems, Inc. MS 6-05 930 DeGuigne Dr. Sunnyvale, CA 94086 +1 408 524-7421 rogerk@mips.COM {ames,decwrl,pyramid}!mips!rogerk "I'm the NLA" "The problem with the rat race is even if you win you're still a rat." - Tomlin