[techtalk] Installing Linux on a second hard drive.

[Lothan]

> From: techtalk-admin at linuxchix.org
> [mailto:techtalk-admin at linuxchix.org]On Behalf Of Jeramia Ory
> Sent: Monday, April 17, 2000 12:58 PM
> To: techtalk at linuxchix.org
> Subject: FW: [techtalk] Installing Linux on a second hard drive.
>
>
> > I must confess, I am a bit confused about why swap partitons on opposite
> > drives would boost performance.
>
>     Caveat: this is all how I understand it, and could be
> completely bogus,
> and would welcome someone correcting me on anything I have wrong.
>
>     One cause of a performance hit is when you swap memory in and
> out of the
> partition.  In a single drive system, even thought the swap is on
> a separate
> partition, it can't access two partitions on the same drive at the same
> time, so you lose performance while the drive alternates the
> memory swapping
> process and the other disk i/o processes.  With the swap on a separate
> channel, the system can now do memory swapping concurrently with the other
> processes, thereby speeding up i/o.  I've found system "snappiness"
> correlates well with how well the i/o is tuned.  FMMV.
>
> > And I shouldn't have any LILO issues if they are each a master, correct?
>
>     You shouldn't have LILO issues regardless of master/slave status, as
> long as you install LILO into the mbr of the primary master.

I see three potential areas this could improve:

1. I/O can be performed concurrently on each IDE bus -- assuming your IDE
controller has dual FIFOs. Some "cheap" third-party IDE controllers
implement a single FIFO and can not perform concurrent access to both ports.

2. Fragmentation is less likely to occur for dynamic Windows swap files.
However, you may as well use a static (fixed size) swap file if the
partition is dedicated to that use.

3. The biggest potential savings is likely in seek time. If the swap file is
on the same physical drive, the heads have to seek a potentially substantial
distance between memory swapping and regular disk I/O. Given the rates of
most modern hard drives, this equates to roughly nine milliseconds for
one-third stroke or as much as 30 to 40 milliseconds full-stroke. Compare
this with .008 millisecond access time for typical PC100 memory and you'll
quickly realize how "expensive" swap files can be.

Lothan