[Techtalk] Largest hard drive useable with an Intel Celeron 466

[Maria Blackmore]

On Thu, 19 Dec 2002, Alvin Goats wrote:

> > I'm thinking of building a fileserver using an Intel Celeron 466 I have.
> > What's the largest hard drive that I could use with that processor (or,
> > if the answer is "depends", how do I find out)?

The easiest way to find out is look up what sort of IDE controller is on
the motherboard, and then research its limitations.

> The CPU is not where the issue resides, although it and the motherboard
> dictate what speed the bus interface is and subsequently what devices
> can be optimally added to the system performance wise. 

This isn't strictly speaking true, the PCI bus will always be either 33 or
66 MHz and either 32 or 64 bits wide.  This gives four possible speed
combinations for PCI, and given a little bit of clever design it's
possible to design a PCI card that will work on any of the four
combinations.

Due to the existance of bridges, a PCI bus will functionally be the same
all the way from an aged 486 motherboard to the latest and greatest dual
Athlon motherboard.

presumably at some point we will also start to see the long promised PCI-X
in things that we can actually buy.

This does of course mean that you can buy an ultra-modern PCI IDE
controller card (eg Promise, Highpoint, etc), and attach the new hard
drive to that instead.  I believe that some of the cards have a
sufficiently smart BIOS on them that LILO can use BIOS int 0x13 to boot
from systems installed on drives on these cards directly.

<snip>

> The original IDE interface couldn't handle more than 504M hard drives,
> the EIDE (or ATA-2) added LBA (logical block addressing) which broke
> most of the memory barriers.

"memory barriers" ?

size limitations I think would be a more appropriate term, sorry to
nitpick, but I can feel my teeth being eroded when I hear people talking
about memory and drive space (semi) interchangably :)

> You probably have an EIDE interface with an enhanced DMA (UDMA/33 or
> maybe a UDMA/66). The UDMA is simply the interface maximum burst mode
> speed. The interface is only a transfer point, addressing (or
> capacity) is done inside the hard drive (the controller chip is on the
> drive and not in the controller card or motherboard).

This isn't strictly speaking true any more.

The IDE interface controller on an expansion card or motherboard has to
translate between things in the machines world, and the world that exists
on the IDE bus, which means that there is a limiting point here too.  This
is defined by the latest standard that this controller supports.

The reasons for this not being strictly true nowadays is that the IDE bus
was originally intended to be an extension of the ISA bus off the
motherboard to the hard drive.  It was intended to present this in an easy
to interface with, cheap and convenient way, and so the original IDE
controller did very little except remove the more complicated and
unneccesary parts of the ISA bus, and buffer it to the IDE bus.

This is why it's called "IDE" - Integrated Drive Electronics

As I said, things have moved on a lot since then, with many additions
having been made.

Of course, in the true PC fashion many of these additions have occured in
a series of dreadful bodges (though not as heinous as those that lead us
from the 8088 to the Pentium 4 :)

They're not all bodges, some are true extensions, but there are those such
as ATA which throw away the original (portion) of IDE.  In the case of
ATA the command set is replaced with something much nicer that's been
borrowed from SCSI.

> There are sometimes some issues with running a higher level UDMA device
> on a lesser one (timing issues with the bus interface) which should be
> looked at with the disk drive manufacturer (UDMA/100 drive on a UDMA/33
> bus).

hmmmm, have to say that I don't remember hearing about this.  The drive
and controller should (99% of the time) be able to arrange to both talk
something that the other is capable of understanding.  Even if it means
dropping back to PIO

> There are modes and PIO's with the EIDE interface that had some
> limits, but if you have a UDMA/33 or higher, you shouldn't have any
> issues with running really large drives. 

PIO means "Polled Input Output"

Horribly slow and CPU intensive, this requires the system driving the
.. uh .. drive to keep prodding at the drive to see what it's doing, and
if it's ready for the next little bit of data, or if it's deigned to spit
out another bit of data yet.

DMA means Direct Memory Access.

Much nicer, the drive and controller arrange to write the data or read the
data from a predefined section of memory, to which access has been
allowed.  Much faster, things can happen in the background without the CPU
having to get involved.

> Ancient MFM, RLL and ESDI drives were "dumb" drives, the controller chip
> is in the card or motherboard and the limit was at the board. Again, the
> board issue was really the controller chip. 

Ah, the good times. When I were a lad^H^H^Hgal we never had any of this.  
we had to carry the data in bit bucks.  Uphill! in the snow! over broken
glass!  Both ways too, mind! AAAND our bit buckets weren't nearly as big.  
You younguns, you don't know how good you've got it, data that transfers
itself! whatever next?!

*ahem* :)

> SCSI controllers have similar issues for disk drives as MFM/RLL/ESDI.

What issue is this then?

> My Adaptec AHA-2940AU has a limit of an 8GB disk, and a maximum of 7
> drives.

Hmm, the 8 GB drive limit was surpassed a very long time ago, and without
breaking backwards compatibility either :)

> Capacity limits vary depending on the interface (narrow or wide) as
> well as the chip set.

hum, capacity limits for SCSI depend on which variant of the SCSI standard
is being used, or to be more exact, which version.  This isn't always the
same as the interface type, and has very little to do with what chipset is
driving the bus, sorry.

*pause to imagine a 6 foot tall surface mount chip behind the wheel of the
280 bus she caught to Oxford last night*

I need more sleep.

Anyway, SCSI has come a long way and a lot further since then.  Massive
drive capacities now being possible, SCSI busses that can have 15 devices
attached and transfer 320 MB/sec across cables which are many metres long
(I believe 12m is the current sensible max).  Being a mere pauper I live
in the comparative ghetto of Ultra160 with a set of IBM Ultrastars, I
dream to move to the heady heights of Ultra320 and FC-AL (Fibre Channel
Arbitrated Loop)

> Firewire currently has a 200GB drive being offered by Western Digital,
> so the capacities are much higher.

Which is actually an IDE drive and a firewire <> IDE convertor :)

> And the firewire can daisy chain up to 62 devices vs. the 7 I have.
> Hmmmmm ...... 200GB x 62 devices ... 12.4 TB? mmmmmmmmmmmmmmm

I wouldn't like to try using that though, doesn't sound very pleasent

> Now, if you want to go to extremes ..... ;):   

One of these http://web6.scan.co.uk/Products/Info.asp?WPID=2222

60 of these http://www.dabs.com/products/prod-info.asp?quicklinx=25CQ

and something to put them in, not to mention a large power supply, and a
chunky aircon unit

8808 GB or 8.6 TB
not bad

could always buy one of these
http://www.netapp.com/products/filer/enterprise-filers.html
:)

24 TB

> EIDE motherboard chips can typically handle 4 drives, assuming one is a
> CD, then you have 3 drives to use.

or add one of these http://web6.scan.co.uk/Products/Info.asp?WPID=17435
or similar

> You can add additional USB drives (assuming your Celeron card has a
> USB interface) and add as many drives as the bus allows.

eww eww eww

<snip much sickness>

Anyway, back to the original subject, if you're still with me :)

I would add a cheap IDE card to your celeron machine.  They're
extraordinarily cheap nowadays, and you've immediately bypassed any
limitation that your motherboard would have had with any large drives.  If
you already have one drive in the machine that you're booting it off, you
won't even need a fancy one with a BIOS.  If you get one with two
channels, you can add two completely seperate identical drives (ok, so you
can do this with one channel, but then it's possible for one drive to kill
the other) and mirror them with software RAID for fast, efficient, cost
effective and reliable storage.

Having said this, I would always personally prefer SCSI, but when all is
said and done, most people don't need the advantages it offers, or want to
pay the premium to get it.

Good luck and stuff

Maria

-- 
"Meow" said Maria the Wonder Cat