*BSD News Article 74629


Return to BSD News archive

Newsgroups: comp.os.linux.hardware,comp.os.ms-windows.win95.misc,comp.os.os2.setup.misc,comp.unix.bsd.freebsd.misc
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.ecn.uoknor.edu!news.ysu.edu!usenet.ins.cwru.edu!agate!howland.reston.ans.net!spool.mu.edu!news.sgi.com!news.msfc.nasa.gov!newsfeed.internetmci.com!news.compuserve.com!ix.netcom.com!netcom.com!stephenk
From: stephenk@netcom.com (Stephen Knilans)
Subject: Re: HELP: Can I mix memory speeds
Message-ID: <stephenkDv2n7u.327@netcom.com>
Organization: Netcom Online Communications Services (408-241-9760 login: guest)
References: <4sr0bg$4ae@uriah.heep.sax.de> <stephenkDuxI2x.B5M@netcom.com> <4t0ur6$7bn@uriah.heep.sax.de>
Date: Wed, 24 Jul 1996 23:43:06 GMT
Lines: 73
Sender: stephenk@netcom15.netcom.com
Xref: euryale.cc.adfa.oz.au comp.os.linux.hardware:45534 comp.os.ms-windows.win95.misc:163039 comp.os.os2.setup.misc:17653 comp.unix.bsd.freebsd.misc:24352

In article <4t0ur6$7bn@uriah.heep.sax.de> joerg_wunsch@uriah.heep.sax.de (Joerg Wunsch) writes:
>stephenk@netcom.com (Stephen Knilans) wrote:
>
>> >Btw., panic does even try to flush the disk buffers, so its effect is
>> >not as desastrous as you describe unless the disk subsystem itself is
>> >hosed.  But in this case, you lose anyway.
>> 
>> Are you saying that Linux intercepts that interrupt, and flushes?
>
>I haven't looked into Linux' code for it, but i would expect it to do
>so.  The interrupt is not ``intercepted'', it's actually being
>delivered as a hardware trap just like all other traps caused by
>hardware events.  It has to be handled by the operating system itself
>in any case.
>
>The usual handling is to panic(), and this usually involves flushing
>the disk buffers.  Of course, this is _always_ a tradeoff between the
>risk of flushing wrong data, and the intention to save as much data as
>you can.  But i think the general consensus is that the latter is much
>more likely useful than the former is to be afraid of.  (If your data
>are not yet there, you are not thinking of what you might have done if
>they were there but were faulty. ;-)
>
>-- 
Although I understand what you are saying, and can agree in part, here is a
summary of what I said, in a DETAILED format.

Parity is NOT a panacea no matter WHAT anyone will say.  It is merely a check
bit that says something MIGHT be wrong with the memory.  Obviously, since it
is only one bit, it can't even do THAT 100% right 100% of the time.  A bad
parity bit means that something is wrong with that bit, or the parity check
hardware, in which case the memory may be GOOD, or that the rest of the memory
is bad or there is a problem with refresh.

This means, simply stated, and this is the ONLY thing it CAN mean, that THE
MEMORY CAN'T BE TRUSTED!  Thus, if it causes a panic to sync the disk, the
program could be corrupted(and thus do ANYTHING), the structure data could be
corrupted(destroying the disk or other sections), or the data could be 
corrupted(which on an encrypted or compressed disk could destroy it).  The one
thing that is certain is that nothing is!  You'd probably STILL want to go to
a backup to avoid checking what might be GIGABYTES of data.  Of course, the
disk structure IS checked to SOME degree by FSCHK, although IT has problems.
SOME data is checksummed, and encrypted or compressed data can be QUICKLY 
verified.  

Frankly, given the above, Parity isn't worth looking all over the planet for.
Of course, today, parity pentium systems can be had for really NO increase in
cost.  If you think it helps, GET IT!  MY advice, however, would be to get 
ECC INSTEAD!  IT can better detect errors, and correct some of them!


So far, every parity problem I have seen was due to pushed hardware, or bad
refresh.  The pushed hardware HAS caused corrupted disks, and the bad refresh
generally just crashes.  Of course, memory can go bad, but other things ARE
more likely.  

As for the earlier crack made on this group about my wanting an award for
rememberring to zip up my fly?  I'll wait until very few do it.  Maybe THEN
I would think it a nice idea. 8-)   Out of EVERY STORE, "field tech", that I 
have seen, NONE used a UPS!  IN FACT, UPS's were VERY rare last time I checked!
Most "UPS" are REALLY STANDBY power supplies.  Of course, NONE of them even
used THAT!  I met 3-4 DEC field techs that knew about static sensitive 
components, and only one store.  Some others actually LAUGHED at the prospect!
Then again, at the last store, where they tried to sell me an earthlink account,
they claimed that there was NO SUCH THING as an IP address, and laughed at the
prospect of having your own web server!

Steve