info-vax@sri-kl (10/24/85)
From: dual!greipa!paul@UCB-VAX.Berkeley.EDU (Paul A. Vixie) > From dual!ucbvax!ulysses!aluxp!wg (Bill Gieske) > > By the way, I now am responsible for 8 VAXes, so am familiar with your > present environment. I don't know how you have DECnet set up on each system, > but I would imagine it is different, i.e. running off different data bases. > Is one node configured differently, such as one is an end node and the other > is a router? This will generate more overhead (I think) on the routing node. > Your figures showed that the performance was the same on both systems be- > fore DECnet was brought up. Well, the figures got even worse when my application was brought up, so I did not suspect DECNet. Silly me. > From (Martin Ewing) > >> My real complaint is that two (supposedly) identical systems >> running identical software take different amounts of time to run the same >> benchmark. >DEC types will tell you that +/- N% execution speed is to be expected >among examples of the same cpu model. N is around 5-10%. There might >be microcode revisions, etc. Nope. Microcode boards were at the same ECO. Good idea, though. >> So I swapped all the CPU boards between the two. All 16 of them. >Oh, no! Oh Yes! >> In a few minutes I will start swapping memory and/or unibus adaptors... >Who handles your insurance? How much is 10% cpu speed worth in repair >bills? I'm glad there are some gutsy hackers left, but all-in-all I'm >just as happy if they don't work for me! I've only met one competent CE from DEC, and zero from CDC. Granted that these guys are "safe": it doesn't cost anything when they drop a board on the floor - this leads to some real attitude problems and a general lack of ability among "company men". Just the same, if I were the manager at this place, the keys to the CPU cabinet would be locked away somewhere, never to be touched except by a "safe" CE... I figure I could yell at DEC until they got someone competent on the contract. But I'm not. And the people who are in charge cut out the service contract in Jan85 when the company was running out of $$$ - we have since been bought by Lotus (the 1-2-3 people), but getting another service contract doesn't seem to be a high priority for them. So when things break, I fix them. You ought to see my spare parts collection... (BTW, I dropped a CDC FSD-160 on it's pointy little head a few weeks back, and we had to pay CDC $100/hour T&M to fix it. When will these people learn? I don't know, but warnings from me don't seem to help...) >P.S. While you're at it, why not measure the systems with the Fast and >Slow clock options? Well, I didn't try that. I am sorta curious, though... I did finally find and fix the problem. It was DECNet... As I said in the original article, I recently imaged one disk onto another as the fastest way to install VMS 4.1 on the backup system. Both systems were therefore routers, looking to the same devices as DECNet lines. But DMF-0 on the backup system was not connected to anything. When I finally SET LINE DMF-0 STATE OFF and SET CIRC DMF-0 STATE OFF, the problem dis- appeared. The lesson is obvious... Oh well, "next time!" Paul Vixie uucp: ...!ucbvax!decwrl!greipa!paul arpa: decwrl!greipa!paul@ucbvax