Re: Memory Reliability

[Schuh, Richard]

It is possible to have a memory error that only affects one user. However, that is rare, almost as rare as having one that affects the entire system. Some of IBM’s memory chips has the ability to correct 1, 2, 3, and 4 bit errors and even an entire memory chip. See http://www-5.ibm.com/hu/termekismertetok/xseries/dn/chipkill.pdf. I would surmise that the only ones who see memory errors anymore are those who read the EREP reports and decide that a card needs to be replaced.   Regards, Richard Schuh   -----Original Message----- From: The IBM z/VM Operating System [mailto:[EMAIL PROTECTED]On Behalf Of Colin Allinson Sent: Wednesday, May 10, 2006 9:38 AM To: IBMVM@LISTSERV.UARK.EDU Subject: Re: Memory Reliability   Thanks for the feedback. We don't actually have a problem at the moment but we are going through operations automation exercise at the moment and all sorts of 'what-if' questions are being asked. My reaction to this one was 'If the once in 30 years event happens then we will have to deal with it as an exception' but I thought it might be good to canvass the opinion of those wiser than I. Regards,   Colin Allinson David Boyes <[EMAIL PROTECTED]> wrote: I haven’t seen any problems with “genuine IBM” memory, but we did have a problem with some 3rd party RAM on our first 9121 that we eventually tracked down to bus speed mismatch between the 9121 and the modules in the memory card.  In that one case, the symptom affected the entire VM system and showed up as corrupted page table pointers for paging operations (which made CP *very* unhappy – flaming death in all directions).  That was a long time ago, and turned me off 3rd party RAM for a long while; it was a bear to track down, and (of course) it was all VM’s fault, since the MVS machine didn’t have those problems (and didn’t have the cheaper RAM either….sigh). Of course, we got the “we told you so” lecture from IBM, too – but they also promptly rebid the memory upgrade to match the price of the 3rd party RAM, so I suppose it was a net win (other than to my sleep cycle).   I’ve never seen a memory error that affected only one user. Have you checked the integrity of your paging areas on disk? I have seen a paging area overlapped by a minidisk that clobbered pages for a rarely used guest (the machine affected got started at IPL, and then went to sleep until needed. Someone did a off-by-one error on a minidisk definition and overlaid one cylinder of page space with a minidisk – when CP tried to page stuff  in for that guest, it got gibberish and the guest promptly died horribly).