magorian@UMD5.UMD.EDU (Dan Magorian) (12/09/87)
I got a call from Kinetics today that they have found a misplaced capacator that causes random Fastpath crashes on noisy Ethernet lines. Apparently Microsoft's 40+ boxes were crashing several times per day, and they demanded a fix. Ours have been locking up as well, but I had put it down to failure to reboot on power dropouts (maybe twice per month). So they will be sending out a letter explaining the patch (a simple solder job, apparently) or people can send them back for free (I was not to use the word "recall"). However, getting the new PROMs is probably a good idea as well. I will suggest that the letter be posted to this mailing list as well. Dan Magorian (301) 454-6030 Computer Science Center magorian@umd5.umd.edu University of Maryland magorian@umdd.bitnet College Park, MD 20742 Applelink: A0190
tim@kinetics.UUCP (Tim McCreery) (12/13/87)
I have not been keeping up with reading this newsgroup lately, but the recent item by Dan Magorian concerning FastPath definitely got my attention. Mostly, I want to clear up a few points with some facts. The "hardware bug", i. e. "misplaced capacator (sic)" is not quite correct. In different network environments, tranceivers, tranceiver cables, etc. inject different levels of noise into the gateway. There is a crystal oscillator circuit to the SIA (usually the AMD 7992) and the sensitivity to noise of the TCLK input to that SIA is also quite variable. Therefore, different units experience noise problems to lesser (i. e. not at all) or greater extent. Reduced noise sensitivity can be obtained by moving a capacitor closer to the serial interface chip, thereby reducing some of the common ground path near it. All current and future units from Kinetics are reducing noise sensitivity in this fashion. We recommend this capacitor movement change to all sites which are experiencing an intermittent failure problem which shows up as the boxes appear to hang (no RTMP packets are being sent), but then fully recover after a power cycle. I should note that most problems reported to us are not like this, but are due to configuration inconsistencies and this "fix" will not solve those problems. With respect to Microsoft, the implication in Dan's article is that ALL the units were failing several times per day. This is false. Because of the size of the Microsoft installation and their excellent network administration, they became an ideal site for diagnosing the problem as being unit and environment specific. We are grateful for their patience and assistance in helping us first document and then find this problem. We are in the process of writing a technical note concerning this problem and the fix and will be mailing it to all of our customers within the next week or so. tim
jim@b-mrda (12/17/87)
I'll vouch for the fix. Of the 12+ boxes that I "control" we were having a problem of them freezing up. sometimes once every other week to 3 times a day. The capacitor is moved from its current location and soldered to the ethernet transciever chip. Since we had our boxes changed not one single one has crashed. jim sadler 206-656-5422 hpubvwa!b-mrda!jim P.O. Box 3707 MS 6R-24 Seattle, Wa. USA 98124 Any opinions expressed are mine and mine only and not that of my employer. Also add in whatever else should be said at this point.
MER@PSUVM.BITNET (Mary Ramsey) (12/23/87)
I got a call from Kinetics today that they have found a misplaced capacator
that causes random Fastpath crashes on noisy Ethernet lines. Apparently
Microsoft's 40+ boxes were crashing several times per day, and they demanded a
fix. Ours have been locking up as well, but I had put it down to failure to
reboot on power dropouts (maybe twice per month). So they will be sending
out a letter explaining the patch (a simple solder job, apparently) or people
can send them back for free (I was not to use the word "recall"). However,
getting the new PROMs is probably a good idea as well. I will suggest that
the letter be posted to this mailing list as well.
Dan Magorian (301) 454-6030
Computer Science Center magorian@umd5.umd.edu
University of Maryland magorian@umdd.bitnet
College Park, MD 20742 Applelink: A0190