Return to BSD News archive
Newsgroups: comp.os.386bsd.development Path: sserve!newshost.anu.edu.au!munnari.oz.au!constellation!paladin.american.edu!howland.reston.ans.net!europa.eng.gtefsd.com!uunet!super!becker From: becker@super.org (Donald J. Becker) Subject: Re: [FreeBSD 1.0R] DMA Problems? Message-ID: <1994Jan5.163110.6408@super.org> Sender: news@super.org (USENET News System) Nntp-Posting-Host: descartes Organization: IDA Supercomputing Research Center References: <CHCErs.G5w@genesis.nred.ma.us> <2encotINN3sq@bonnie.sax.de> <CIDrrv.3MJ@sneaky.lonestar.org> <jmonroyCJ5JwH.EzE@netcom.com> Date: Wed, 5 Jan 1994 16:31:10 GMT Lines: 36 In article <jmonroyCJ5JwH.EzE@netcom.com>, Jesus Monroy Jr <jmonroy@netcom.com> wrote: [[ Typical useless JMonroy drivel deleted. Why doesn't he just admit that some SCSI controllers are known DMA bus hogs. Oh well, it doesn't matter, it's not like he's every going to write code. ]] > You most certainly should; John knows what he's talking about. It's >interesting about the cosmic ray stuff - cosmic rays are a major source of >alpha particles. In the early days of DRAM designs, alpha particles were the >primary failure mode for DRAM - they were such a problem that nobody would >seriously consider a DRAM design that didn't implement ECC. The 256k DRAMs were >where the problem really became serious - some engineers were sure that >smaller micron technologies would be impossible because of the alpha particle >problems. ...But then there was improved die 'coating' materials that >dramtically reduced the problem. Cosmic rays aren't the only cause of alpha >particles, in fact I think that the plastic package that contains the die is >actually a greater source of them then cosmic rays (but I'm fuzzy on this). It's been quite a while (15 years?) but I seem to remember that the "cosmic ray" problem was really alpha particles from *ceramic* packages. (The ceramic being used at the time was slightly radioactive.) This was being mistaken for cosmic ray bit upsets -- well known from earlier space/rad-hard work. The claim was that DRAMs larger than 64K were impossible because shrinking the capacitor would result in a single hit taking out the stored bit. Further, the cascade effect of a cosmic ray being slowed just above the die and taking out multiple would make even ECC difficult. Thus today we are stuck using large boxes full of 64K DRAMs backed by bubble memory. -- Donald Becker becker@super.org IDA Supercomputing Research Center 17100 Science Drive, Bowie MD 20715 301-805-7482