OT: redhat

[Bill Vermillion]

On Sun, Nov 14 00:19 , Men gasped, women fainted, and small children 
were reduced to tears as Bill Campbell confessed to all:" 

> On Sun, Nov 14, 2004, Bill Vermillion wrote:
> >On Sat, Nov 13 15:01 , Bill Campbell, showing utter disregard for 

> ...

> >Remember the FAT12.  And remember when you could only have
> >about 512 files total.  [I think it was 512 - it may have been
> >slightly different].  THen came 16 bit fat and of course the 2GB
> >limits were there.  So you made three 2GB partitions, and then
> >took whatever was left and made as many logical partitions to use
> >up the drive space.

> Microsoft's first ``solution'' to large file systems was to
> crank up the segment size which increased the total size
> without increasing the number of segments. One effect was
> to waste large amounts of disk space because of inefficient
> storage.

And their first solution to the 512 ?? file limit was to create
PC-DOS 2.0 [this was before they came with the MS-DOS line with
their own impramatur.]  When I first read about that it looked like
they had answered all the problems I had with DOS 1.x.  You could
almost call that the first solution to large file systems and it
was sorely needed.  Mountain had come out with a $2500 add-on
box that let you add a 5MB HD to the system - and without the 2.0
and the directory hierarchy those would have been useless.  Since
they were reworking the structure they had an opportunity to do it
right but almost got the feeling the MS people didn't understand
the base OS they had bought and modified.

After running it for about 6 months I gave up.  I found that my old
8-bit systems were faster on anything that had screen output
because of what I considered the stupid way they hanlded screen
output.  Everthing I had used up until them used memory-mapped
video.  The only way the PCs would approach the same speed
was to turn off any output to the display.   

So at the end of 6 months I was running two systems - my old 8-bit
and a Radio Shack 16 with Xenix.  That was late 1983.  


> >So many of those limits are gone now.  Remember when the OS for
> >booting had to like within the first 1024 cylinders.  When you look
> >at the current FreeBSD booting limits you wonder how we survived so
> >long with the old limits.  In the 5.x and 6.x FreeBSD the booting
> >code has to lie withing the first 1.5 Terabytes of the first
> >filesystem.  The first filesystem does not have to be limited to
> >that amount, just the the boot code must like in the first 1.5TB of
> >it.

> The grub boot loader was the first Linux loader to allow
> booting partitions beyond the first 1024 cylinders with LILO
> not far behind.

> >But there is the possible problem of losing a drive in a spanned
> >system.  But with the 400GB drives now available that should only
> >be a problem on really large systems, and then you will most likely
> >use a RAID system as a spanned FS with no RAID recovery could be
> >a long time to restore - even with the fast systems.

> RAID has its own problems, mostly related to OS driver
> support (unless one uses purely hardware RAID as in extern
> RAID controllers that connect to standard SCSI controllers
> independent of the underlying Operating System). Fundamentally
> you run into the problem of ``putting all your eggs in one
> basket'' when dealing with spanned file systems. RAID 5 and
> it's descendants permit one to replace failed drives without
> data loss, but the concatenated file systems don't. How much
> data do you want to have to recover in the case of a hard drive
> failure?

That's why running filesystems that span disks seems to be risky if
you aren't running a RAID environment.  And I keep seeing problems
in NLs with people running SW RAID solutions.  It only takes one
serious crash to realize that their cheaper solution [SW RAID]
was more expensive in the end.

> >That makes sense.  Does Linux still change the numbers if you
> >remove a drive in the middle - say you had 4 drives and removed one
> >- do the drives above the middle one get renumbered.  That used to
> >be a problem but I'm not that deep into Linux at the moment.

> I'm not sure what Linux does when a partition is removed, and I really
> didn't want to find out the hard way which is why I use LVM.

Some OSes have the habit of renaming the HDs on boot if something
disappears.  You need to hardwire the controllers to device
names to avoid that.  

Sort of like the MS way of giving new letters when you put in 
another HD or device, and see everything fail because D: is no long
what it once was.   But it they hadn't done that where would all
the companies like Partition magic be :-)

Bill
-- 
Bill Vermillion - bv @ wjv . com