[Sluglug] Re: Hard drive access optimization
Ignacio Solis
isolis at igso.net
Wed Oct 5 15:41:20 PDT 2005
* Thomas Leavitt (thomas at thomasleavitt.org) said:
> ... aren't there advantages, in terms of access speed, to having all the
> data towards the inside of the disk (heads don't have to move as
> far?)... I always thought that was the primary reason a defragger moved
> everything to the "front" part of a partition. ...and one of the reasons
> that people pay lots of money for high performance SCSI drives and then
> stripe data across them, but don't fill them up....
One of the key factors on how a drive performs is the filesystem and how/when
the metadata is stored.
In the case of FAT, the metadata is at the beginning of the disk, hence having
the data there will mean the head will have to move less between data and
meta-data.
The optimal solution is to have most of the data/meta-data in the center of the
disk. Writing outwards/inwards (farther away from the center in both
directions) as the data is less frequently accessed. The problem is calculating
this access probability.
Modern filesystems try to keep the meta-data close to the data.
New drives might be faster on the outside of the disk. This is due to the fact
that they rotate at the same speed, and at the outer edge of the disk the head
goes over more area than close to the center of the disk. Older disks divided
tracks into sectors of equal angular length, current ones vary depending on the
distance from the center.
The reason you stripe data and it's faster is because you can theoretically
read and write at n-times the speed, where n is the number of disks.
SCSI disks are faster for multiple (small?) reasons. Actuators, settling time,
rotational speed, command processing, bus, etc.
Nacho
--
Homepage: http://www.igso.net/~isolis/ | EEE8 08C9 FBAE B471 9691
GPG Public Key: http://www.igso.net/isolis.gpg | CE7A 1CC8 D3DE B31E 10AB
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://sluglug.ucsc.edu/pipermail/sluglug/attachments/20051005/e51f06e2/attachment-0002.pgp
More information about the Sluglug
mailing list