ron@brl-tgr.ARPA (Ron Natalie <ron>) (06/01/84)
Received: from brl-vgr.arpa by BRL-TGR.ARPA id a018837; 1 Jun 84 12:41 EDT Date: Fri, 1 Jun 84 12:39:36 EDT From: ron@BRL-VGR.ARPA Subject: Subject: followup failed Sender: ron@BRL-VGR.ARPA To: ron@BRL-VGR.ARPA inews: No title. Your article follows: Newsgroups: net.periphs Subject: Re: FPS maintenance questions References: <440@islenet.UUCP> We were left with a bad opinion of FPS (or as we refer to them: Fly By Night Convolvers) after our sole purchase of an AP-120B four years ago. The thing came with installation included. They installed it in the rack, but were not able to run checkout because even though we had specified that we were running a PDP-11/70 they had no software that would work on the processor. For the first few months whenever someone tried to use it under load the system would die with UNIBUS errors. This was when I came to work for BRL. Looking at the UNIX software that we had bought from another vendor (Pennisula Research (not the WICAT people, the Stanford ones)), it stated that there is a defect in the UNIBUS interface that prevents it from working on the faster PDP-11's and that you should get the fixes that they put in the Stanford system. I called FPA and talked to their hardware department. I told them that our system had never worked, not been tested on installation, and explained the note about the ECO's done to the Stanford system. I asked if they would tell me what the ECO's were. They admitted that there were several ECO's that were needed to make that revision board work with anything other than PDP-11/34's, but wouldn't tell me what they were. I kept calling different levels of customer support, but the answer was always the same. I could either buy a new interface board for $930 or I could pay transportation + hourly rate for one of their service engineers to come put them in. I told them that since it never worked that I sould be entitled to a new board, but was willing to settle for just a description of the ECO's. No dice. Finally, I tracked down one of the Stanford guys at his new job and got the ECO information from him and fixed the board myself. No longer hung the UNIBUS. Then one of our programmers tried a little benchmark to invert a 30 by 30 matrix and multiply it by the inverse to get a identity matrix. On the 11/70 in BASIC-PLUS it took something like 24 minutes. On the AP-120B, it took 17 seconds. Unfortunately the answer from the array proccessor was wrong. I mentioned to the programmer that there was a speed/accuracy trademark that he was going to have to make. The pattern of the error was rather regular and the Penninsula Research diagnostics would now run and found a bad memory board, which we also fixed locally and got the beast running. The total is that we got a processor with a broken UNIBUS interface and one broken memory board, and paid for installation that never really got done and FPS was totally unwilling to provide even minimal assistance to get us out of the mess. -Ron