Plans for hot adding and removing memory [LWN.net]

Plans for hot adding and removing memory

Posted Jun 17, 2013 20:34 UTC (Mon) by etienne (guest, #25256)
In reply to: Plans for hot adding and removing memory by dlang
Parent article: Plans for hot adding and removing memory

Because he has ECC RAM he needs to write it once to initialise the ECC bits.
Well, theoretically no memory should be read before being written, but in practice someone may do it (obviously in another OS, never happen on Linux).
In linux, if you get an ECC error, you directly assume the RAM is faulty, not that some area is read before being written.

to post comments

Plans for hot adding and removing memory

Posted Jun 17, 2013 20:52 UTC (Mon) by dlang (guest, #313) [Link] (1 responses)

right, but even on servers with 128G of ram that I have, this only takes a few seconds (and I suspect that a noticeable chunk of the time is spent updating the display to report progress :-)

nowhere near the several minute figures mentioned above.

A full memory check on these 128G systems does take a few minutes.

Plans for hot adding and removing memory

Posted Jun 18, 2013 12:44 UTC (Tue) by nix (subscriber, #2304) [Link]

It's not doing a memory check, that takes ages and I turned it off.

But it is quite true that BIOSes are so opaque (and so badly written) that it could very well be spending its time doing something else, probably terribly inefficiently!

But, still, the 'need to write everything' and the ever reducing ratio of memory bandwidth to RAM volume *is* eventually going to have the effects I suggest above, even if it isn't now. So it's good that Linux already has the machinery necessary to fix it, if the BIOSes would please catch up. (Actually, looking at the article more closely it had this from 3.2 or thereabouts, since all this needs is plugging, not unplugging.)