Slashdot Mirror
Tracking Down a Single-Bit RAM Error
Hanji writes "We have discussed here before the potential effects of and protections against cosmic ray radiation, but for the average computer user, it's an obscure threat that doesn't affect them in any real way. Well, here's a blog post that describes a strange segfault and, after extensive debugging, traces it down to a single bit flip, probably caused by a stray cosmic ray. Lots of helpful descriptions of Linux debugging techniques in this one, and a pretty clear demonstration that this can be a real problem. I know I'm never buying a desktop without ECC RAM ever again!" The author acknowledges that it might not have been a cosmic ray-based error, but the troubleshooting steps are interesting no matter what the cause.
One of my computers had an intermittent failure in a RAM chip/line/something somewhere that mostly manifested as SHA/MD5 failures when I was checksumming large files that I'd downloaded. Never showed up in Memtest86, but eventually I eliminated every other possibility. IIRC, I solved it by underclocking the machine and then replacing it when I was able.
When I was in college one of my physics professors told us he doubted programs would ever get bigger than a few hundred kilobytes because cosmic rays would cause the larger programs to fail too frequently.
I don't know about cosmic rays, but immediately following the Easter day Earthquake in Guadalupe Victoria (about three hundred miles from where I was located) I tried to fire up my laptop and then my desktop, both of which had been suspended to RAM. Neither one would wake up, though the lappie displayed a garbled screen. No errors in the log files (Ubuntu 9.10 on the sys76 lappie, Deb Lenny on desktop).
Forget a RAM error, I have seen a bit on a file on the disk flip.
After years of successful operation a Perl script quite working. On investigation a G was transformed to a W a difference of one bit. The file mod date was years old.
Soft errors in DRAM are far more likely to be the result of alpha particle decay from materials in the die and packaging.
Would it really be so hard to read the article before posting?
"Prefiero morir de pie que vivir siempre arrodillado!"
I've been working with some large microarray datasets recently, and so had to double my computer's memory to 8GB.
As I've done for years, I went to Fry's to get some Corsair chips... installed F13 64bit to replace my older 32bit distro... and crash-o-matic began. Mostly from Chrome and Mercurial.
I ran memtester86+ and sure enough, verified my first purchase of faulty memory.
So, I went back to Fry's and exchanged for another pair of Corsair 2GB chips. This time, I ran memtester86+ first thing... ANOTHER bad set, so back it sent to Fry's.
*Third* set of memory was Kingston, and a trip through memtester86+ verified no errors. Yay!
Computer has been stable, too.
With more and more RAM in computers, my next box will have ECC.
Its interesting to me because my first instinct would have been to assume something got corrupted and my first step would have been to reboot. If the problem persisted through a reboot then I might have gone down the rabbit hole in similiar
fashion to try and find and fix the root cause.
There are enough sofware bugs, kernel bugs, driver bugs, hardware hiccups due to marginal equipment, power fluctuation, interference, random noise... and i suppose even cosmic radiation that I would rarely think to spend the time to trace a transient problem unless it was reproducible accross reboots, or at least happened on multiple separate occasions.
I would think it's more likely there is trace radioactive elements in the epoxy the chip is encapsulated in.
Actually, I recall reading that in the early solid state memory days, they had problems with this. I don't remember what the solution was, but I thought it was to make the circuit somewhat resilient to it, as it was impossible to get 100% neutral epoxy, there's always going to be traces of something radioactive.
I think they tested the cosmic ray theory by running the same chip with and without lead shielding, and did not find a significant difference in errors, they then assumed it was impurities in the chips themselves decaying.
Sent from my PDP-11
Back in the early 80's, HP published a paper on random bit errors in RAM. They looked at chips from a variety of vendors and determined that the RAM coming out of Japan was the most reliable. That paper caused a lot of US RAM vendors to shutter their doors as there was a sea change in purchasing habits.
A few years later, I ran into John Scully while we were waiting for a flight. I mentioned the paper to him and asked him how Apple could seriously expect to sell a Macintosh specifically aimed at the Scientific community if it didn't have ECC. He blithely said "it's not a problem..." 20+ years hence and most of us still don't have ECC so it seems he was right.
I'm putting tinfoil hats on all of my servers, right away!
This is one area where AMD is light years ahead of Intel. With Intel, you have to buy a Xeon and a server chipset to have ECC support, which basically is going to run you at least a grand or two just for the CPU and motherboard (at least if you want an i7 based Xeon). AMD on the other hand supports ECC across the board, and you just need a motherboard which supports it, which is most of them (total cost: <$500).
Thanks for the gouging Intel!
Game! - Where the stick is mightier than the sword!
And no doubt in these super high transistor count and clock frequency CPUs and chips we are using these days there must be devices and methods used inside them to keep the logic transfer and computation validity on the straight and narrow.
Other than ECC on the cache arrays... No. Not a scrap.
If you want reliability on every internal signal and register against cosmic ray strikes, because you're a military or aerospace contractor, you pay boku bucks for it, settle for having way less than what we would currently call performance. And even then I highly doubt anyone is actually putting ECC on each and every bus or set of latches. You just radiation harden the device as much as possible, and then use three of them so if one gets the wrong answer because of a particle strike, the other two will out-vote it.
The enemies of Democracy are
Disks have a lot, and I mean a LOT of ECC on them. It is not a situation of "I need to write a 1 so I'll place one at this location on the drive." They use a complex encoding scheme so that bit errors on the disk don't yield data errors to the user.
Then there's the fact that bits aren't even stored as bits really. All current drives use (E)PRML which is (Enhanced) Partial Response Maximum Likelihood. What this means is bits aren't encoded as a high-low state or FM wave or any of that. They are written using flux reversals, but the level is not carefully controlled, it can't be. So when you read the data the drive actually looks at an analogue wave. It encodes the partial response it gets, and then finds the maximumly likely pattern that matches.
Sounds like voodoo but works really well. Things are not simple thresholds or the like, it is a complex system and ends up being quite robust and resilient to error.
So it is highly unlikely that you had a bit flipped on a disk. Would require some amazing circumstances to happen. The RAM error is far more likely. Not just the cosmic ray thing but, as the parent noted, bad RAM. Normally when RAM fails, it fails catastrophically and it is immediately apparent. Not always though. It can not only fail on single bit locations, but only during certian ops. That is why memtest does so many different tests. One kind might works fine, another might fail. Rare, but I've seen it on a few systems.
My RAM is shielded against cosmic rays by my mothers basement.
You are on the right track. As someone with over a quarter century of background in combined embedded software and hardware design (the most recent decade for life-dependant systems), it always amazes me how quickly pseudo-technical people jump to wild speculation for observations that they cannot explain.
They fail to understand that a hardware system is an imperfect representation of the theory (probably the biggest failure in the schooling of software developers and even some hardware is to get this message into their heads). While they feel comfort in the theory of a binary system, they utterly fail to understand that our real systems, like us, are imperfect and, like us, live in an analog world. Simple things like temperature variations, noise from common (rather than cosmic) sources, marginal design timing, imperfect components, simple intermittents, etc., are 10^24 times more likely the cause.
But they're not as fascinating as wild speculation, are they?
I had a mysql replication server which was reading SQL commands from a binary log on a master server. One day after years of operation I noticed an update failed. I didn't see anything at first by looking at the query, but when I looked closely I noticed the query had a single character changed, and of that character only one bit had changed. It was something like a P becoming a Q and thus giving a syntax error.
True story.
The worst problem was with ceramic DIP packages -- the really good ones for when you needed reliability (partly because the plastic ones tended to allow moisture to get in, and then condensation on thermal cycling.) The standard ceramic packaging material contained trace amounts of thorium, which is an alpha emitter. The alpha bombardment was enough to flip bits.
There have been several fixes since then. Using materials that don't contain radioactive species was one. The one you're probably remembering is that the manufacturers apply a polymer coating to the surface of the die, which is enough to stop a lot of alpha particles and a fair number of electrons. Getting rid of lead in packaging is also good, because lead tends to contain some radioactive traces.
On the other hand, there's flat nothing to be done about cosmic rays and damn little to be done about X-rays and thermal noise (you do keep your memory cold, don't you? Thermal noise is proportional to KT/qe after all.) So at some point we get to where there are too many bits which need minimal energy to flip them -- and then you have errors.
Pity that so few mobos actually support ECC, though.
Lacking <sarcasm> tags,
I am both shocked and amazed that you eventually broke up.
I feel fantastic, and I'm still alive.
Billions of years in the ground, and only a few centuries on the roof and all of the radioactivity is gone! Wow!
The author needs to provide a reference, but there's a few ways I can think of that a processing stage, and a few centuries would produce something less radioactive than something produced more recently. I think all of them stem from the ore containing a source material that gets separated through the refining process, but the daughter products from the source don't. Here's one scenario:
Ore = Lead + radio-isotope a + radio-isotope B.
radio-isotope A decays to radio-isotope B
radio-isotope A: 4 billion year half-life.
radio-isotope B: 20 year half life, decays to stable isotope C.
during refining, radio isotope A gets nearly completely refined out to parts per trillion. radio isotope B is similar to lead chemically, and remains at 1 parts per million (at time of refining).
200 years go by. (10 half lives of radio isotope B)
radio isotope B is now at 1/2^10 concentration, or about 1 part per trillion. Significantly less than when it was first refined. The added radioactivity from radio-isotope A decaying into B is negligible due to the long half-life of A.
These numbers and process are obviously made up to show how it MIGHT work. It still remains to be seen if it's actually true or not.
AccountKiller
Walla!
https://www.eff.org/https-everywhere
The guy that posted this is a Ksplice developer. In case you didn't knew, KSplice allows you to patch your running kernel without rebooting. Nice.
Anyway, this guys sees a random memory error. He conveniently goes on a debugging rampage, while we all know the most logical first step would be rebooting that damn machine. Random memory errors do happen.
He says he "hasn't gotten around" to memtesting his RAM yet. So, let me get this straight ... he implies that random cosmic rays caused the error, but he hasn't yet tested his ram for what is the most possible cause of the issue?
Then he goes on to explain that you don't even need to reboot your machine due to damn cosmic radiation. Or kernel updates. Because you have Ksplice.
Come on.
WTF am I doing replying to an AC at 5 A.M on a Friday night?
On the subject of the imperfect nature of machines, I found this post by Richard D. James (aka Aphex Twin, a noted electronic music composer) quite interesting. He describes how the physical machinery of analog electronic music machines means it is near impossible to duplicate them in digital programs.
link
Author: analord
Date: 02-07-05 03:14
some people bought the analogue equipment when it was unfashionable and very cheap though.
some of us are over 30 you know!
anyone remember when 303`s were £50? and coke was 16p a tin? crisps 5p
also you have overlooked A LOT of other points because its not all about the overall frequency response of the recording system its how the sound gets there in the first place.
here are some things which you can`t get from a plugin,they are often emulated but due to their hugely complex nature are always pretty crass aproximations..
the sound of analogue equpiment including EQ, changes very noticably over even a few hours due to temperature changes within a circuit.
Anyone who has tried to make tracs on a few analogue synths and make them stay in tune can tell you this,you leave a trac running for a few hours come back and think Im sure I didnt fucking write that,I must be going mental!
this affects all the components in a synth/EQ in an almost infinte amount of tiny ways.
and the amount differs from circuit to circuit depending on the design.
the interaction of different channels and their respective signals with an analogue mixer are very complex,EQ,dynamics....
any fx, analogue or digital that are plugged into it all have their own special complex characteristics and all interact with each other differently and change depending on their routing.
Nobody that ive heard of has even begun to start emulating analogue mixer circuitry in software,just the aesthetics,it will come but im sure it will be a crap half hearted effort like most pretend synth plugins are.
they should be called PST synths, P for pretend not virtual.
Every piece of outboard gear has its own sound ,reverbs,modulation effects etc
real room reverb, this in itself companies have spent decades trying to emulate and not even got close in my opinion, even the best attempts like Quantec and EMT only scratch the surface.
analogue EQ is currently impossible in theory to be emulated digitally,quite intense maths shit involed in this if youre really that interested,you could look it up...good luck.
your soundcard will always make things sound like its come from THAT soundcard..they ALL impose their different sound characteristics onto whatever comes out of them they are far from being totally neutral devices.
all the components of a circuit like resistors and capacitors subtley differ from each other depending on their quality but even the most high quality milatary spec ones are never EXACTLY the same.
no two analogue synths can ever be built exactly the same,there are tiny human/automated errors in building the circuits,tweaking the trimpots for example which is usually done manually in a lot of analogue shit.
just compare the sound of 2 808 drum machines next to each other and you will see what I mean,you always thought an 808 was an 808 right?
same goes for 303`s they all sound subltey different,different voltage scaling of the oscillator is usually quite noticable.
VST plugins are restricted by a finite number of calculations per second these factors are WAY beyond their CURRENT capability.
Then there is the question of the physicallity of the instrument this affects the way a human will emotionally interact with it and therfore affect what they will actually do with it! often overlooked from the maths heads,this is probably the biggest factor I think.
for example the smell of analogue stuff as well as the look of it puts y
A few years ago I came across a thread on a FreeBSD mailing list where a build of some package was failing and the submitter couldn't tell why because he wasn't a developer. The failure was unusual and no one else could reproduce it. Eventually, the problem was traced back to a character in the source differing from the original. The character was a one-bit difference from the correct character, and it was suggested to the submitter that he reboot and memtest his memory. Sure enough, one single bad bit out of around 512MB.
The really impressive thing is that this guy resisted the urge to just reboot his machine. Otherwise, the clues would have vanished and the expr binary would have run again without any issue.
Maybe that's why the first step one takes when something behaves weird on a Windows system is to reboot it...
Uh, no, not at all... *shifty eyes*
The enemies of Democracy are
Roman ingots to shield particle detector
http://www.nature.com/news/2010/100415/full/news.2010.186.html
It was me.
Sorry 'bout that.
You live below your mother's basement???
Sure. In his mother's sub-basement.
"I don't care about the Constitution!" --Bill O'Reilly, November 17, 2009
I know, cosmic rays sound so much cooler, but it's far more likely he has some crappy memory and/or his memory refresh timings are too high.
DRAM memory cells have to be refreshed pretty often (anywhere from 7.8usec-12usec), otherwise they become unreliable. If his BIOS has the memory timings set to something obscurely long, it may be there are specific rows/cells on his DRAM modules that are too weak to read after bleeding off a bit of charge. Changing the refresh timing would likely improve the situation, causing the memory to refresh it's state more often.
I shouldn't have spent all my mod points yesterday. I guess my hardware knowledge is obsolete; I had no idea modern HDDs don't store individual bits anymore.
"I don't care about the Constitution!" --Bill O'Reilly, November 17, 2009
And we used to blame Microsoft engineering team for all the crashes we experienced !!
Although interesting, TFA it is without a doubt the most pedantic and roundabout way I've ever read of establishing your rig is not stable.
From TFSA:
And in fact, since that incident, I've had several other, similar problems. I haven't gotten around to memtesting my machine, but that does suggest I might just have a bad RAM chip on my hands.
Yeah he has a stuck or semi-stuck bit and a hour or two of his life he won't get back.
In such a circumstance I've found underclocking and overvolting the DIMM might coax it to work again but it's best to RMA or bin it.
After logging in slashdot still does not take you back to the page you were on. It's been that way for 20 years.
Checking at alternate.de (not the cheapest online shop, but good for comparisons because they have both consumer and server parts):
Intel:
The only Socket 775 boards that support ECC seem to be those with the 32xx MCH chipset. Starting at 195 Euros (Asus P5BV-C).
For Socket 1156, the consumer chipsets allow ECC but you still need to find a board with BIOS support. Sadly Alternate does not list the ECC support status, but you might find one that supports ECC among the cheaper ones for 80-90 Euros. You do, however, need a Xeon which starts at 213 Euros (Xeon X3430, 4 x 2.4 GHz)
So mainboard plus a quad CPU costs you around 300 Euros at Alternate.
AMD:
Board situation (Socket AM3) similar to Intel's Socket 1156, boards with ECC support are available for 80-90 Euros.
Unlike Intel, even cheap desktop CPUs support ECC. As a cheap quad, Alternate offers the Athlon II X4 635 for 108 Euros.
So mainboard plus quad CPU costs you around 200 Euros, 100 less than with Intel.
C - the footgun of programming languages
Woohoo!
While working as a failure analysis technician at a company that made a disk controller, I came across a single-bit error in static RAM cache that was repeatable. I was lucky to have the software and hardware tools available and I eventually tracked down the failure mode. Setting a bit at a certain location would cause another, different location's bit to get set. Just that one bit. And only if you set it. Resetting it did not cause the other bit to reset.
This turned out to be a manufacturing problem with a particular run of RAM. I starting finding more of these bad parts and could reproduce the failures. I guess what I'm saying is that this could well be a manufacturing defect in the RAM.
Best regards.