RAM question for everyone!
gilboad at gmail.com
Fri Jan 23 16:38:42 UTC 2009
On Fri, 2009-01-23 at 16:26 +0000, Bryn M. Reeves wrote:
> Alan Evans wrote:
> > On Fri, Jan 23, 2009 at 7:43 AM, Dan Track <dan.track at gmail.com> wrote:
> >> I was recently asked a question about how much RAM should there be
> >> within a server given that the APP uses 8GB of Memory, should I buy
> >> 10Gig of memory and have a small harddrive and no swap space? Would
> >> this configuration allow everything in my OS to run from RAM and not
> >> from swap? If this is the case then there's no need to ever create
> >> swap, is there?!?
> >> Your thoughts are appreciated.
> > This question, along with other recent discussion about swap, leads me
> > to ask a question in response: Why is everyone so concerned about how
> > to get away without swap?
> > Hard drives are cheap. Why does your server with potentially 10GB
> > (!!!) of RAM have a hard drive so small that you can't sacrifice a few
> > GB for swap?
> I think many people aren't as concerned about sacrificing a bit of
> disk space as much as they are concerned about the performance impacts
> when the system begins to use the swap, especially for desktops.
> Linux will attempt to move old data that has not been referenced for
> some time out to the swap device even when there is relatively little
> pressure to do so. This is generally a win since we are better
> utilising the physical memory of the system (storing more
> frequently/recently used data in it) but it may lead to nasty delays
> when the swapped-out data is needed again.
> This is more of a problem today than 15 years ago because of the ever
> widening gulf between main memory speeds and (HD based) mass storage
> speeds (or at least, seek times).
> As an example, try opening something in OpenOffice and then minimizing
> it for a week. Even if the box was fairly quiet for that period,
> chances are that much of OO's address space is now swapped out.
> Clicking the window in the task bar will cause the system to churn for
> a few seconds or more before the app returns to a usable state.
Yeah, but this problem can more-or-less be avoided by
More information about the users