Date
1 - 5 of 5
Importance of ECC RAM on the build computer
@gimballock
Is there generally suggested advice regarding the importance of ECC on the build computer? I'm very new to Yocto so I'm not sure if there is a chance the final build image could experience some corruption due to flipped bits in RAM or if I'm totally overthinking this. It is also possible that Yocto does some kind of CRC checking of its own that makes this a non-issue. Any guidance would be appreciated. Thank you!
Yoann Congal
Hi!
Le ven. 17 févr. 2023 à 06:08, <davidskeck@...> a écrit :
Is there generally suggested advice regarding the importance of ECC on the build computer? I'm very new to Yocto so I'm not sure if there is a chance the final build image could experience some corruption due to flipped bits in RAM or if I'm totally overthinking this.
Last time I checked. The ECC vs non-ECC price difference was not worth the time you may spend in case of a RAM bitflip.
Here, we put ECC RAM in our build systems and I guess it's quite the industry standard.
It is also possible that Yocto does some kind of CRC checking of its own that makes this a non-issue. Any guidance would be appreciated. Thank you!
Yocto does source integrity checks (the stuff it downloads from the internet) but it can't do much more after that.
I've yet to debug a RAM bitflip but I guess it may trigger any kind of weird behavior which may be hard to debug and time consuming.
My advice is : keep your time and sanity, use ECC RAM.
Regards,
Yoann Congal
Smile ECS - Expert technique
Alexander Kanavin
On the other hand, I'm yet to hear an actual case of a broken build
due to non-ECC. It's not impossible, but it seems exceedingly rare.
You're far more likely to spend 100x as much time on logical errors or
out-of-resources problems.
Alex
toggle quoted message
Show quoted text
due to non-ECC. It's not impossible, but it seems exceedingly rare.
You're far more likely to spend 100x as much time on logical errors or
out-of-resources problems.
Alex
On Fri, 17 Feb 2023 at 09:31, Yoann Congal <yoann.congal@...> wrote:
Hi!
Le ven. 17 févr. 2023 à 06:08, <davidskeck@...> a écrit :
Is there generally suggested advice regarding the importance of ECC on the build computer? I'm very new to Yocto so I'm not sure if there is a chance the final build image could experience some corruption due to flipped bits in RAM or if I'm totally overthinking this.
Last time I checked. The ECC vs non-ECC price difference was not worth the time you may spend in case of a RAM bitflip.
Here, we put ECC RAM in our build systems and I guess it's quite the industry standard.It is also possible that Yocto does some kind of CRC checking of its own that makes this a non-issue. Any guidance would be appreciated. Thank you!
Yocto does source integrity checks (the stuff it downloads from the internet) but it can't do much more after that.
I've yet to debug a RAM bitflip but I guess it may trigger any kind of weird behavior which may be hard to debug and time consuming.
My advice is : keep your time and sanity, use ECC RAM.
Regards,
--
Yoann Congal
Smile ECS - Expert technique
Alexandre Belloni
On 17/02/2023 10:41:37+0100, Alexander Kanavin wrote:
final binary which was then not working as expected.
We also had multiple cases of headers/source file corruption after the
tarball is extracted but usually those are either caught by the compiler
or don't matter at all.
--
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
On the other hand, I'm yet to hear an actual case of a broken buildIt actually happened in one of our build, were we had a bitflip in the
due to non-ECC. It's not impossible, but it seems exceedingly rare.
You're far more likely to spend 100x as much time on logical errors or
out-of-resources problems.
final binary which was then not working as expected.
We also had multiple cases of headers/source file corruption after the
tarball is extracted but usually those are either caught by the compiler
or don't matter at all.
Alex
On Fri, 17 Feb 2023 at 09:31, Yoann Congal <yoann.congal@...> wrote:
Hi!
Le ven. 17 févr. 2023 à 06:08, <davidskeck@...> a écrit :
Is there generally suggested advice regarding the importance of ECC on the build computer? I'm very new to Yocto so I'm not sure if there is a chance the final build image could experience some corruption due to flipped bits in RAM or if I'm totally overthinking this.
Last time I checked. The ECC vs non-ECC price difference was not worth the time you may spend in case of a RAM bitflip.
Here, we put ECC RAM in our build systems and I guess it's quite the industry standard.It is also possible that Yocto does some kind of CRC checking of its own that makes this a non-issue. Any guidance would be appreciated. Thank you!
Yocto does source integrity checks (the stuff it downloads from the internet) but it can't do much more after that.
I've yet to debug a RAM bitflip but I guess it may trigger any kind of weird behavior which may be hard to debug and time consuming.
My advice is : keep your time and sanity, use ECC RAM.
Regards,
--
Yoann Congal
Smile ECS - Expert technique
--
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
Leon Woestenberg
Hello,
Interesting question, and I especially would like to hear on the MTBF
*measured* through the ECC detection algorithm. It can usually detect
more errors than it can correct, and I wonder if the systems admin are
notified when bitflips where detected.
Regards,
Leon.
--
Leon Woestenberg
leon@...
T: +31 40 711 42 76
M: +31 6 472 30 372
Sidebranch Embedded Systems
Eindhoven, The Netherlands
http://www.sidebranch.com
On Fri, Feb 17, 2023 at 3:14 PM Alexandre Belloni via
lists.yoctoproject.org
<alexandre.belloni=bootlin.com@...> wrote:
toggle quoted message
Show quoted text
Interesting question, and I especially would like to hear on the MTBF
*measured* through the ECC detection algorithm. It can usually detect
more errors than it can correct, and I wonder if the systems admin are
notified when bitflips where detected.
Regards,
Leon.
--
Leon Woestenberg
leon@...
T: +31 40 711 42 76
M: +31 6 472 30 372
Sidebranch Embedded Systems
Eindhoven, The Netherlands
http://www.sidebranch.com
On Fri, Feb 17, 2023 at 3:14 PM Alexandre Belloni via
lists.yoctoproject.org
<alexandre.belloni=bootlin.com@...> wrote:
On 17/02/2023 10:41:37+0100, Alexander Kanavin wrote:On the other hand, I'm yet to hear an actual case of a broken buildIt actually happened in one of our build, were we had a bitflip in the
due to non-ECC. It's not impossible, but it seems exceedingly rare.
You're far more likely to spend 100x as much time on logical errors or
out-of-resources problems.
final binary which was then not working as expected.
We also had multiple cases of headers/source file corruption after the
tarball is extracted but usually those are either caught by the compiler
or don't matter at all.
Alex
On Fri, 17 Feb 2023 at 09:31, Yoann Congal <yoann.congal@...> wrote:
Hi!
Le ven. 17 févr. 2023 à 06:08, <davidskeck@...> a écrit :
Is there generally suggested advice regarding the importance of ECC on the build computer? I'm very new to Yocto so I'm not sure if there is a chance the final build image could experience some corruption due to flipped bits in RAM or if I'm totally overthinking this.
Last time I checked. The ECC vs non-ECC price difference was not worth the time you may spend in case of a RAM bitflip.
Here, we put ECC RAM in our build systems and I guess it's quite the industry standard.It is also possible that Yocto does some kind of CRC checking of its own that makes this a non-issue. Any guidance would be appreciated. Thank you!
Yocto does source integrity checks (the stuff it downloads from the internet) but it can't do much more after that.
I've yet to debug a RAM bitflip but I guess it may trigger any kind of weird behavior which may be hard to debug and time consuming.
My advice is : keep your time and sanity, use ECC RAM.
Regards,
--
Yoann Congal
Smile ECS - Expert technique
--
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com