samba - Feb 1999 - OT ECC Memory (WAS: Large uptime)

Granted ECC should be more reliable over the long term, but how long is
that?

I could buy "luck" if it was one machine, but we're talking about
two
machines in my house, and at least one at a customer site, and they're ALL a
couple of years old at least.  (Mine are 486/80's, and the customer has a
fairly old Pentium.)  The year-and-a-half one (which I admit may not have
been real because I didn't have any contact with them for most of that time)
was an old 386-25 I installed in 1990 that got pulled out about 4 years ago.

The Pentium at the customer site runs SCO Open Server and rarely gets
rebooted unless I go in and do it.  I was there last weekend and did it for
the first time in MONTHS.

The guy I work with has a lot of Netware installations that also have very
long uptimes.

Just based on real-world observation, I find those bit-failure rates a
little hard to swallow.

Next time I speak to my associate I'll ask him about this and see what his
experience has been.  He does Windoze and Netware almost exclusively so he
deals with a lot more of this kind of stuff than I do.
> -----Original Message-----
> From:	Denis Sbragion [SMTP:sbragion.denis@usa.net]
> Sent:	Thursday, February 25, 1999 3:00 AM
> To:	KeHall@exchange.ml.com; Multiple recipients of list
> Subject:	RE: OT ECC Memory (WAS: Large uptime)
> 
> Hello,
> 
> At 04.20 24/02/99 +1100, Hall, Ken (ECCS) wrote:
> >I have Linux and Netware servers that have run well over 6 months at a
> time
> >with no problems.  Not only don't they have ECC, I don't even
think the
> >SIMMS have parity!
> 
> I think this may be luck. I also had one customer that run a Linux server
> without ECC for few months with no problem, but it was just one. I admit
> also I saw the "1 bit a month" report few years ago. May be
memory chips
> became more reliable in the meantime. It was also an average report, so
> some memory chips may provide better performance. Anyway it's sure that
> DRAM chips loose bits, else why do ECC memory exists? Supposing new memory
> chips loose just 1 bit a year per megabit on the average, multiply this by
> the typycal 32 or 64 MBytes of ram of a small server and you'll get the
> picture. I *always* put a saturday night reboot in the crontab of non ECC
> servers, whether they are Linux or not. Furthermore, just to get the full
> picture, I *never* saw a problem on a Linux server whith ECC.
> 
> Bye!
> 
> 	Sbragion Denis
> 	InfoTecna
> 	Tel, Fax: +39 039 2324054
> 	URL: http://space.tin.it/internet/dsbragio

Hello,

At 01.49 26/02/99 +1100, Hall, Ken (ECCS) wrote:
>Granted ECC should be more reliable over the long term, but how long is
>that?
...
>Next time I speak to my associate I'll ask him about this and see what
his
>experience has been.  He does Windoze and Netware almost exclusively so he
>deals with a lot more of this kind of stuff than I do.
well, I believe you. May be it's the opposite of what I said, i.e. I've
been unluck and the memory I used was more faulty than the average. Anyway
it's sure I had problems getting long uptimes on all the servers without
ECC I installed, except one. This means 8 servers out of 9 (8 Linux, 1 NT
SP4). May be I've been *really* unluck. This things can be tracked only
with good statistics, and 9 servers isn't a good one. The data I have
anyway seems to agree with the '1 bit a month report'... uuuhhhmmm,
things
going a little crazy here! Anybody out there has reliable data about this
subject? I know it's completely off topic but I think server reliability is
an important issue for Samba.

Bye!

	Sbragion Denis
	InfoTecna
	Tel, Fax: +39 039 2324054
	URL: http://space.tin.it/internet/dsbragio