🚀 go-pugleaf

RetroBBS NetNews Server

Inspired by RockSolid Light RIP Retro Guy

Article View: comp.sys.ibm.pc.hardware.storage
Article #331

Re: Fastest portion of a HD....

#331
From: c1saxton
Date: Fri, 16 Feb 1996 15:04
78 lines
4985 bytes
In <4fk04r$o7l@steel.interlog.com>, ldh@interlog.com (Laurent-David Hasson) writes:
>
>I have read somewhere that there was actually quite a difference in performance 
>depending on where data is on a disk. From what i understood, near he center of 
>the disk, transfer rates and access times are quite higher than if the data were 
>at the edge of the disk platter. Is that correct, is it the opposite, is it 
>drive dependent, or is it just plain untrue?
>
>Which ever way it is, when you partition a disk, does the partitioning layout 
>start from the center of the disk platter or near the edge?

1.  Modern drives put more data in the tracks near the edge so the outermost
    cylinders give better performance.

2.  Partitions start at the outermost cylinders.

3.  Although point 1 gives some advantage to outer cylinders, the biggest gain
    in performance can be had by intelligent partitioning and distribution of files.

Ideally, you want all your frequently-accessed files clustered together over a
set of contiguous cylinders so that the read/write heads only ever have to move
a tiny amount.  Unfortunately, three things conspire to thwart that design.

1.  Rarely do we know, and even more rarely do we know in advance, what
    files we need to access most frequently.  We may have some knowledge for
    our own data files but not for secondary files used by programs and operating
    systems.

2.  The installation procedures for operating systems and application programs
     give us very limited control over the location of their various components -
     frequently-used and rarely-used files are interspersed with abandon.

3.  The disk and directory structure of older operating systems constrains our
     ability to assign our files to optimal locations.

DOS and its descendants install only on the primary partition of a drive (is that
still true for Windows '95?) which perforce is the outermost partition.  If you
put programs and data on other partitions then it means that switching between
system files or directories and user files or directories forces a large seek.

There are good reasons for installing the operating system in its own partition,
keeping installed applications and data separate and insulated from system
upgrades but even if you attempt to minimise seek times having a single
partition containing everything the DOS file system works against you.  The
root directory and the file allocation table are at the beginning of the disk so
seeks to the outer cylinders are common.  If you are running Windows then the
swap file is probably near the beginning of the disk as well.

Given the above, you'd probably be better off using two or more physical drives.
Put a smallish C: partition on the first drive for the operating system.  The second
physical drive would hold all your frequently-used stuff and any spare space on
first drive could be partitioned off to hold stuff which is hardly ever used.

You have more freedom with modern operating systems such as OS/2, and to a
slightly lesser extent Windows NT.  When I only had one drive I had a 300Mb C:
drive containing WinOS2 and stuff which I used occasionally, a 100Mb D: drive
containing OS/2 and not much else, a 300Mb E: drive containing the swap file and
heavily-used programs such as the Internet Access Kit, utilities, compilers, editors
and so on.  The remaining 300Mb F: partition was used for rarely-used stuff;
typically things downloaded and/or archived in .zip format and my source code
repository.

Allocating stuff that way meant that almost all the time I was working near
the middle of the disk.  Formatting all partitions to HPFS rather than FAT meant
better caching algorithms and space usage and resistance to file fragmentation.
In addition, HPFS puts the directories in the middle of the partition instead of at
the beginning which implies, on the average, shorter intra-partition seeks.

I don't know if Windows 95 offers that much flexibility but if so the I would urge
you to consider a similar layout.  If you typically do just one thing at a time then
it probably doesn't matter a damn but if you surf the Web, run C++ compiles
and play Galactic Civilizations during the wait times, you'll get far more benefit
from organising your files than depending on extra sectors on the outer cylinders
of your disk.

--
Jon Saxton, NYPC Developers' SIG coordinator
Inhabiting an OS/2 oasis in a desert of VM sand
Technical Details
Message-ID: <4g26d3$gim@watnews2.watson.ibm.com>
Path: arc1-ca.pugleaf.net!watnews.watson.ibm.com!usenet
References: <4fk04r$o7l@steel.interlog.com>

Show Full Headers
Complete Article Headers
Path: arc1-ca.pugleaf.net!watnews.watson.ibm.com!usenet
From: c1saxton@watson.ibm.com
Newsgroups: comp.sys.ibm.pc.hardware.storage,comp.periphs.scsi
Subject: Re: Fastest portion of a HD....
Date: 16 Feb 1996 15:04:03 GMT
Organization: IBM T.J. Watson Research Center
Message-ID: <4g26d3$gim@watnews2.watson.ibm.com>
References: <4fk04r$o7l@steel.interlog.com>
Reply-To: c1saxton@watson.ibm.com
NNTP-Posting-Host: guido7.watson.ibm.com
X-Newsreader: IBM NewsReader/2 v1.2.5