[Beowulf] Re: ECC Memory and Job Failures (Huw Lynes)

[Lux, James P]

> -----Original Message-----
> From: beowulf-bounces at beowulf.org 
> [mailto:beowulf-bounces at beowulf.org] On Behalf Of Vincent Diepeveen
> Sent: Monday, April 27, 2009 7:31 AM
> To: Robert G. Brown
> Cc: Beowulf Mailing List
> Subject: Re: [Beowulf] Re: ECC Memory and Job Failures (Huw Lynes)
> 
> 
> 
> 
> On Apr 24, 2009, at 4:20 AM, Robert G. Brown wrote:
> 
> > On Thu, 23 Apr 2009, John Hearns wrote:
> >
> Yeah usually it is a hardware or software bug.
> 
> However i noticed that if i saved my tapes (Ultrium2-200GB 
> uncompressed, 400GB according to the salesman) at about 1 
> meter above the ground here, things go fine. Yet if i save 
> them at 2nd floor here that is more risky.
> There is definitely more bitflips there. Now that is a lot 
> closer to the 2 x 450 MVA powerlines, which are on average at 
> a height of about
> 19 meters above ground,
> a bit less when a lot of power is on 'em, as the cables tend 
> to get very hot then.
> 
> Now on paper these powerlines are supposed to not radiate the 
> ionising radiation very far. Say half a meter max (which 
> doesn't really matter for life & death, as at 2 meters 
> distance you're already dead in this wet climate).
> 
> Would this measurement prove that ionising particles, which 
> cause of course cancer, sometimes travel further away from 
> these cables?
> 

There are so many physics things wrong with this, it's hard to know where to start.

1) tapes store data as magnetic domains, so charged particles aren't likely to be an issue: it's not like a semiconductor memory or gate where the energy from the charged particle deposited in the gate flips a bit or discharges/charges a capacitor.

2) magnetic fields from the power lines are possibly an issue, except that the field at your position is so far below the level required to flip the bit that I'd think it's unlikely. Several issues: the power lines are almost a group of 3 phases, where the net field from all is pretty much cancelled at any reasonable distance.  As an exercise, I gave a colleagues undergrad E&M statics class the challenge of telling whether you could use a handheld magnetic compass to detect the field under the 4GW DC transmission line near here. (two wires, carrying 4000A at 1MV). Short answer is, only if the system is operating with one wire disconnected, and using the sea return, so it's unbalanced. That's 4000A about 20m above you, with a sensor that can detect a field on the order of 1E-5T or 1E-6T (earth field is 1E-4 T).  

3) EM fields (at least at the levels you're talking about) are not ionizing (clearly, the field at the conductor is high enough to ionize the air so you get corona discharge). And the radiated fields are certainly not ionizing. The photon energy at 50Hz is so low that there is no known chemical or biological reaction with an activation energy low enough that 50Hz photons could make it happen. There have been invitro studies that showed some effects from magnetic/electric fields at power line frequencies, but it's a long way from some small change in cells in culture medium to carcinogen.

If you drag out the IEEE/ANSI standard for EM exposure (IEEE C95.1-2005) (which has been harmonized with equivalent EU standards), there is probably over a hundred pages of summary of over 1000 papers in the literature on this. (all thanks to media hype and poor statistical analysis in books like "currents of death" and popular media poorly reporting on the famous Kaiser VDT study)