BERTRAND Joël wrote:
Hermann Lauer a écrit :On Sun, Sep 14, 2014 at 07:44:31PM +0200, BERTRAND Jo??l wrote:hmm well where will this info usually show up? in the system/kernellogs? if it's not showing up then that means the freezes must be causedby something other than ECC errors?On console and system logs. Freezes can be caused by kernel bugs and there are a lot of kernel bugs in sparc64 port.I have systems with wheezy where I saw some corrected ECC (console and logs).But the main problem are the freezes - there is some discussion on the sparc kernel over freezes under io load. So probably some of that bugs are fixes in the 3.17 rc's > rc1 - will try when time permits.Maybe, but I haven't see this kind of discussion for a long time. I cannot test 3.17-rc1 as sun4v support is broken (LSI SAS and cpu support).
I could potentially test it, but in order for that the happen the Debian team would have to make sure that usable CD images were being built. They can't shift the blame for that upstream.
And don't forget that some chipsets (psycho for example) are buggy. U80 triggers false ECC errors when system runs with 4 GB of RAM.
I was working through spare machines and boards earlier in the year, and found a disturbing number where POST was reporting Psycho failures which broke PCI hence ebus etc.
I'd be interested to hear others' experiences: is Psycho known to have a limited service life?
-- Mark Morgan Lloyd markMLl .AT. telemetry.co .DOT. uk [Opinions above are the author's, not those of his employers or colleagues]