Subject: Re: VIA VP2 chipset
To: None <current-users@NetBSD.ORG, port-i386@NetBSD.ORG>
From: Greg A. Woods <>
List: current-users
Date: 02/04/1998 23:44:54
[ On Mon, February 2, 1998 at 07:57:01 (-0500), David Jones wrote: ]
> Subject: Re: VIA VP2 chipset
> This is unlikely.

Yeah, I know all of that.  I was only kinda joking about an alpha
particle being bigger than multiple memory cells -- we might get there
someday, but I doubt were anywhere near that, esp. in consumer grade

However you have not disputed that DRAMS continue to suffer noticible
error rates that are trivial and inexpensive to protect against.

Unfortunately the PC industry seems so keen on shaving every last penny
off the price of systems no matter what the true cost to the user ends
up being.  Of course this likely means that even if your memory
subsystem is 100% reliable there's still a good chance that errors will
be introduced either on the bus or even within the CPU.

Even worse there's at least one analysis I read recently that suggested
modern DRAMs are have much higher error rates when they are accessed
(read and write) more often.  This seems to be supported in the field by
folks who say that they have fewer problems with modern operating
systems (like "NT" and even "Linux" ;-) when they use ECC because these
modern systems are more susceptible to memory errors.  It's not just
folklore either.

Sure I know of PC based systems that have run reliably for great lengths
of time, even under significant load (eg. a 486-based http, mail, ftp,
etc. server that had an uptime of over 800 days w/NetBSD).  The risk of
someone getting a wrong byte in an e-mail message or web page, or even
an ftp'ed file, is of course much higher and far less noticable than
that of a critical OS data structure being corrupted, and that may be
why many people never "notice" any errors even on long running systems
that use very stable software, but I'd bet they certainly would if they
had reports from their ECC protection circuits.

I'm not suggesting that we be so pedantic as to calculate every move
three times over like NASA does when the lives of astronauts are at
stake, but since we do know that DRAMs are by their very nature unstable
and error prone, and since we do have the technology to detect the most
common form of error they suffer, and even to correct most of them, why
wouldn't we use it even if it costs a few dollars more?????

							Greg A. Woods

+1 416 443-1734			VE3TCP			robohack!woods
Planix, Inc. <>; Secrets Of The Weird <>