[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Linux Il
On Thu, 25 Jun 1998, Ira Abramov wrote:
> On Wed, 24 Jun 1998, Ariel Biener wrote:
>
> > > Your /dev/sda1 is probably the swap partition.
> >
> > And if it is, just for the record, it is VERY bad practice to have the
> > swap partition as the 1st partition.
>
> since it seems to be that the fastest tracks of the disks I used lately
> are in the beginning, I see no harm in that.
>
> funny. up untill 2-3 years ago, the end of the disk was fastest, anyone
> has any ideas when did that switch?
When: no. But why: - The boot sector and other startup and boot-related
parts of an installed OS are the least often used parts of a system. Thus
it makes sense to put them on the inside tracks which are slower. They
also have a slower data rate usually and thus less read errors. They fill
up to 20% of a large disk, thus user land software gets to run faster and
be loaded/saved faster. - A special benefit is, that if the disk crashes
somewhere, any flying debris and dust go OUTWARDS by centrifugal effect.
When the boot sector was outwards, it got some dust, and it was usually
the end of the disk and all of its data. Now, if you crash a disk, it will
seldomly refuse to boot, and register errors in the crashed cylinder and
upwards, in the path of the debris ;), which fly outwards. With this
arrangement it is very unlikely that debris get onto the boot cylinder,
AND the heads are parked far away from it (always at the far end of the
platter).
With some large disks run under diagnostic software you can see this, as
the diag. software can be told which cylinder range to massage, and it
does, and you can see where the heard arm goes for cyl 0. with the cover
open. (don't do this at home).
BTW there are high end disks that do a RAID for the 1st 8-16 tracks
(wherever cyl.0 is, for the width of the head apron ;). I think it's RAID
0 or 1. The user sees nothing, but the chances that you can't access the
boot sector after a hiccup are reduced violently. It is almost impossible
to measure this from user level, unless you measure the WRITE speed for
the 1st N tracks and detect a speed increase I think (but I'm not sure,
because of the buffering in between and the possibility of late write-back
in the drive electronics).
Re: partitioning: imho the Large-Disk-HOWTO explains a lot about this
under Linux.
And last: Does anyone know if Linux is capable to exploit swap partitions
distributed as rings in a single large disk, interleaved with data
partitions (i.e. is the elevator algorythm in new kernels still there, and
does it apply to swap partitions ?).
Peter