[juneau-lug] Re: Space Available on Hard Drive

  • From: Tony <tony.taylor@xxxxxxxxxx>
  • To: juneau-lug@xxxxxxxxxxxxx
  • Date: Fri, 07 Sep 2001 15:10:07 -0800

A few comments here, from the peanut gallery (honey-roasted, of course):

You can use RAID devices under Linux if you use the logical volume
manager (LVM), found at: http://www.sistina.com/products_lvm.htm (a
personal gripe: I hate the MS-influenced 3-letter extension "htm." 
C'mon, people, we're not stuck in the stone.age anymore!  There, got
that out of my system.)

You can do your various RAID levels (0, 1, 3, 5, and combinations of the
above).

However.

I would not suggest RAIDing IDE drives, except for mirroring, and only
then if you mirror two drives on separate channels.  IDE has poor bus
handling; one command per channel is all that's allowed.

SCSI, on the other hand, can issue multiple requests on the same channel
(even to the same drive) simultaneously.  So you can recieve information
from one drive, while issuing a request to another (parallelized
requests).  If performance is an issue, either just run RAID 1
(mirrored/duplexed) on two IDE drives on two channels (ie, one a primary
master, the other a secondary master), or do your striped (RAID 0) array
on a SCSI system.

A good intro to RAID is at:

http://www.acnc.com/04_01_02.html

In any case, I would recommend putting your system on a single drive
independent of the RAID system; otherwise, you may find it difficult to
mount your system for recovery in case of the worst.

Second, use "df -k" instead of df; then the report is in kilobytes, and
not blocks.  On most Linux systems, a block is 1k; but on many other
systems, it's not (say, like on my Solaris boxen, where a block is 512
bytes).

You can change the block size when you cook the filesystem with the -b
option (for ext2, at least).  Check out the man page for mke2fs for more
info, including options of inode count, reserved block counts, etc.


Lastly, make sure you know the size of the blocks on each of your
devices; your filesystem block size should be a multiple of the physical
device's block size-- otherwise, you'll take a huge performance hit. 
Granted, since most devices have either 512-byte or 1k blocks, that
shouldn't be a problem.  (Watch out for hard drives with 2k blocks,
though, and make sure you cook the filesystem with a 2k block, which is
becoming more popular.)

Anyway, just my free advice.  I hope it's worth every penny.

                                - Tony

hans snyder wrote:
> 
> I am sure someone else will come to your rescue, but I -THINK- that I 
> remember hearing that you could stripe multiple
> hard drives togeather and mount one in anothers partition....but I don't 
> remember how...I also seem to remember
> reading that besides NFS (or whatever it is) files system for linux there 
> were at least two others, one by IBM and one
> by someone else...(oracle-no doesn't sound right)? So I think what you are 
> saying is possible...I encourage you to
> keep looking! -hans
>

------------------------------------
This is the Juneau-LUG mailing list.
To unsubscribe, send an e-mail to juneau-lug-request@xxxxxxxxxxxxx with the 
word unsubscribe in the subject header.

Other related posts: