[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Partitioning for Speed



Alvin Oga, 2003-Mar-31 17:03 -0800:
> 
> hi ya

Greetings
 
> On Mon, 31 Mar 2003, Jeff wrote:
> 
> > I've seen a reference to two regarding the location of a partition on
> > the HDD being faster than other parts of the HDD.  I've been trying to
> > get a definitive answer on this and it's still not clear to me.  
> 
> 
> the disk spins at the same rotational rpm ... its the "same" speed

Now that I sit and think about that, I see it now.

> what you can do, is to add more data ( sectors ) on  a given tract
> 	say 64 sectors per track vs 256 sectors on the outer tracks
> 	( doing so screws up your pll and other analog signals 
> 	( but is worth it if you want to maximize capacity
> 
> - whether reading  64* 512bytes/sector   or 256 * 512byte/sector
>   is better/worst is up to the app ... and how data is written

I'll have to ponder this for awhile.  Don't quite get it yet.

> - less moving of the heads is generally faster  as it takes too long
>   to move the head relative to just reading data 
> 
> 
> - track 0 ( first partitions ) is ALWAYS on the outside of the disk

Thanks for clearing that up for me.

> - example pic of partition layout on the disk
> 
>         http://www.linux-1u.net/Partition/

Fantastic web site!  I'll be spending some time there.  Pictures are
good!

> - careful selection of bytes per sector also helps
> 	512 byte /sector  vso 1024 or 2048 bytes/sector
> 	( bytes per inode )
> 
> - but all that is lots-o-headaches 
> 	and you'd need to write data to disk to take those extra
> 	tweeks into account for maximum benefit

Uhuh.
 
> - calculate ... 5400rpm vs 7200rpm vs 10,000rpm vs 15,000rpm
> 	- defaults is 512 bytes/sector
> 	- defaults is 64 sectors per track
> 	- defaults is "n" number of physical heads ( number of platters )
> 	
> 	- for each revolution of disk ... you get xxxKB/sec of disk
> 	transfer
> 
> 	- ignore all those marketing numbers of ata-66, ata-100,
> ata-133
>
> - for ide disks  -- 1st round no-brainer speed comparason/optimization
> 
> 	- fiddle with some of the hdparm options and check its effects
> 
> 		- hdparm -tT  /dev/hda
> 
> 		- hdparm -d 1 -u 1 -m 16 -c 3 /dev/hda
> 		( make sure your disk supports the options first )
> 
> 		- turn on dma ( -d 1 )
> 		- turn on intr ( -u 1 )
> 		- turn on multicount ( -m 16 )
> 		- turn on 32bit io ( -c 3 )
> 
> 		- set the proper ATA speed ( -X69 for ata-100 )
> 
> 	http://www.linux-1u.net/Disks/hdparm.txt

I've spent time learning how to use hdparm and have been able to boost
performance for my drives most of the time.  I haven't gotten anything
better than around 20Mbps, but that's better than 4.

> 
> - dont forget to account for 2MB disk cache and 8MB disk cache
> 
> - simpler to just add a 2nd spindle ( /dev/hdc ) and be able
>   to read data 2x faster than with one disk ( /dev/hda )
> 	- initial writing of data and parity will be slightly slower
> 
> - other ways to improve your system
> 
> 	http://www.linux-1u.net/Tuning/

Thanks Alvin.  I've got some studying to do.  :-)

jc

-- 
Jeff Coppock		Systems Engineer
Diggin' Debian		Admin and User



Reply to: