[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#838919: debian-installer: please calculate swap parition according to max RAM supported by the motherboard



On Monday, September 26, 2016 10:23:02 AM EEST Lennart Sorensen wrote:
> On Mon, Sep 26, 2016 at 05:06:18PM +0300, Martin-Éric Racine wrote:
> > Package: debian-installer
> > Severity: wishlist
> > 
> > As far as I can tell, d-i calculates the size of the swap partition
> > according to the curently installed amount of RAM.
> > 
> > Whenever the RAM is upgraded later on, the swap parition no longer
> > fulfills its intended purpose, in cases when it would be needed to store
> > hibernate images, since the parition was calculated to store a much
> > smaller RAM image.
> > 
> > A more desirable method would be for d-i to use 'dmidecode' to probe the
> > system's memory controller for the maximum amount of RAM that is
> > supported and to calculate the swap partition size according to that.
> Of course many people never upgrade their ram, and often going beyond 50%
> of maximum gets very expensive.
> 
> Also many people never suspend.
> 
> Also dmidecode is x86 only and some older systems didn't have it, so it
> wouldn't help the majority of architectures.
> 
> So while it is an idea, I am not convinced it doesn't actually create
> problems.
> 
> Besides I could move the disk to a new system later which could have
> even more ram, so even making it as large as dmidecode says won't solve
> all cases.  I would think that case is at least as likely as maxing out
> the ram in a machine is.

Besides, some servers (and similarly workstations) have exorbitant memory 
capacities. I've managed some servers with 256GB RAM capacity, and since these 
servers were designed for processing roles, they had ~300GB disks. In that 
case, most of the disk will be wasted as swap which may never be used in its 
lifetime.

So in either desktop or server case, dmidecode way looks more problematic. It 
potentially wastes space, and addresses a rare problem.


Reply to: