[vox-tech] Three Install Questions

Peter Jay Salzman p at dirac.org
Mon Feb 21 08:58:35 PST 2005


On Mon 21 Feb 05,  7:39 AM, Wilson Shealy <wtshealy at comcast.net> said:
> Hi everyone,
> 
> I appreciate all of the great suggestions I've received from this list.  It 
> seemed appropriate that I should follow up with the practical solution to 
> my problem,  
> 
> I never determined what the cause of my boot problem was but I did find a 
> workaround.  I created a boot floppy and then copied the boot sector of the 
> boot floppy into the root directory of my XP install as a binary file.  By 
> pointing the windows installer at this binary file, it will now "boot" into 
> GRUB which will load my Linux install.  
> 
> I don't know why this wasn't working when trying to copy the boot sector 
> directly from the partition.  It may have something to do with the reiser 
> file system but I really am only speculating.
> 
> As for swap space,  I've decided to go with Karsten's suggestion and create 
> multiple swap partitions.  This way I can mount the additional swap after 
> adding more RAM (probably next month).

Hi Wilson,

The "swap is double RAM" (hereafter "SIDR") rule is _really_ outdated.  I
believe it even predates Linux.  At the very least, it predates when *I*
started to use Linux.

SIDR comes from a long, long time ago when memory management on Unix was
much different from today.  Back then, Unix tried very hard to keep a full
image of the RAM layout on hard drive.  This means you had to have AS MUCH
swap space as RAM, but if the kernel wanted to swap a page of resident
memory to disk, your swap space had to be MORE than the amount of physical
RAM.

How much more?

For a server (and Unix back then was synonymous with "server"), you wanted
to be able to swap a whole lot.  So the general rule was "double".  That's
where the SIDR rule comes from.

Today (and I don't know if this was ALWAYS true for Linux, but it certainly
was true when I started to use Linux), memory management strategies are much
different.  Linux attempts to keep as much in physical memory as possible
and resorts to swap space only when it needs more pages of memory.

Note word: "needs".

If your computer isn't using swap to begin with, there is no reason for you
to increase swap.  With memory for home systems relatively cheap these days
if you ever hear or see your computer swapping, go out and buy yourself
another DIMM (if you're using Linux.  Reboot if you're using Windows).

Don't think that your computer will magically need more swap just because
you unscrew the case and stick another DIMM in there.   If you're dealing
with like ... Ultrix from 1987, then yeah.  You'll need to increase swap.
However, the truth of the matter is that when you stick a DIMM into a modern
Linux system, your system will need LESS SWAP, NOT MORE, unless you start
running extra applications just for the heck of it.

So let me restate:

   * A modern Linux system does not need to follow the SIDR rule (and I
      believe that Linux never did need to follow SIDR).

   * Just because you add more RAM does not mean you automatically need
      to increase swap.

Use "cat /proc/swaps" to see how much swap your system uses.  Look at it
every so often to get an idea of what trends look like.

I'm guessing that some Redhat employee must have been an old Unix admin who
was hired by RH to write their user manual.  He probably put that rule into
the RH User's Manual because that's what he always knew and did.  From
there, it probably made its way into a bunch of HOWTOs and books.

One last note about swap:

1. Having multiple swaps per disk is of almost no benefit at all.

2. Having a swap per physical drive (or SCSI bus) is beneficial only when
   the swap partitions have the same priority.

Pete

-- 
Save Star Trek Enterprise from extinction: http://www.saveenterprise.com

GPG Fingerprint: B9F1 6CF3 47C4 7CD8 D33E  70A9 A3B9 1945 67EA 951D


More information about the vox-tech mailing list