Re: filmscanners: Best solution for HD and images

["Ezio c/o TIN" <ezio.lucenti1@tin.it>]

Pat , you are right .
Please, let me add some comments .... while OT ... I think this matter of
efficiency and speed is an argument directly involving our group seen that
e-photography is meaning big amounts of data and (eventually) long waitings
in front of a screen .

>The Adaptec RAID
>card works somewhat differently and allows multiple
>channels to behave as one. Raid 5 then can be thought
>of as n number of disks with the data striped as in
>Raid 0, but with parity error information saved along
>with the data. The amount of redundant data used is
>1/n, so the total amount of storage is (drive
>capacity)*(n-1).

As far as I know this is the definition of RAID 5 with the distribution =
sharing of the so called ''check sum / parity'' byte on the disk chain !
Be careful because in case of multiple chains you will need multiple spare
areas (as many as the chains).
Thus in the case of SCSI where you cannot (by definition) overcome the
number of 6 devices x chain/controller,  you will have , as a minimum ,
100/6 = 12.7% used by the parity and then 87.3% of AVAILABLE space for the
user LESS what it is taken by the stack from the theoretical capacity of the
disks.
BTW , this method compulsorily implies a DOUBLE WRITING need i.e. write the
data + write the new parity (even if on another disk) and this is meaning a
LOSS OF SPEED/EFFICIENCY (relatively to the achievable speed/efficiency of
the disks) relatively to other RAID techniques like RAID 1 .... RAID 0 is
(by definition) the opposite side of the domain i.e. full speed/efficiency
but NOT SAFETY (as you correctly state).

AFAIK, more over , to overcome the ineherent limitation of IDE protocol and
be able to put more than 2 devices x controller , the vendor has to multiply
(internally to the disk controller) the number of IDE controllers and rejoin
everything to the ''image'' of a single virtual controller and so on ....

I really cannot understand why it would be needed such a complication and
dependancy from the controller vendor when the SCSI hard drives cost almost
the same (or 20% more max) of IDE hard drives !
Sometime ago the difference was remarquable and then people not willing to
have large amount of data available at workstation level were trading off
speed + efficiency + safety with a cheap hard drive .... but today ? ... an
IDE RAID (or array) controller costs more/same as a good SCSI controller ,
providing less speed , less functions , less efficiency and less safety .

Is it a matter of principles ?

Yesterday night I have bought on eBay an IBM U-160 10000rpm 18GB new and
under warranty for 102 USD + 20$ of shipment to Italy from USA and this unit
will be the fourth inside my system while having 2 x CD/R (IDE) and 1 x DVD
(IDE) .
My aggregated sustained transfer rate is 3 x 35MB/s + 1 x 29MB/s = 134MB/s =
the limit of a 32bit PCI bus at 133MHz (but still in the limits of an
Adaptec 29160 controller)

Sorry , but I cannot perceive the price/performance advantage of an IDE
solution . Not even in the case of  a lower amount of data requested.

Sincerely.

Ezio

www.lucenti.com  e-photography site


----- Original Message -----
From: "Pat Perez" <patdperez@yahoo.com>
To: <filmscanners@halftone.co.uk>
Sent: Sunday, November 11, 2001 10:00 PM
Subject: RE: filmscanners: Best solution for HD and images


> I'll jump in here.
>
> Raid 0, striping, assigns half of the data to one
> drive, the other half to the other drive. The writes
> happen more or less simultaneuosly, so large file
> operations happen in roughly half the time.
>
> Raid 1 is, as you said, mirroring, where all data are
> duplicated, so that if a drive fails, another exact
> copy is ready to take over (after a command to 'break'
> the mirror set).
>
> Classically, Raid 5 has a single chain logically (IDE
> is limited to 2 devices per chain, and the above have
> both drives on one chain (channel)). The Adaptec RAID
> card works somewhat differently and allows multiple
> channels to behave as one. Raid 5 then can be thought
> of as n number of disks with the data striped as in
> Raid 0, but with parity error information saved along
> with the data. The amount of redundant data used is
> 1/n, so the total amount of storage is (drive
> capacity)*(n-1).
>
> Raid 0+1 is an IDE only hybrod that allows you to use
> both channels with two disks such that you get
> mirroring of two striped disks.
>
> I know of no motherboard based IDE Raid solutions that
> are designed for high performance (Preben mentioned
> that the host CPU does the drive I/O processing).
> Apparently, the Adaptec IDE Raid PCI card handles
> this. I have no experience with this card (though I
> use Adaptec SCSI adapters exclusively). At work I have
> a card similar to what Preben described. It is made by
> 3Ware (www.3ware.com), and is called the Escalade. I
> can heartily recommend this card under Win2000, the
> only OS I have used it with. It is available with
> support for 2, 4, or 8 IDE drives. At this time, it
> doesn't yet support the new ATA133, so I guess you'd
> be limited to 8 of Maxtor's 100 gigabyte drives!
>
> There are actually several other levels of RAID, but
> they are mathematically undesirable as regards the
> cost/benefit/performance tradeoffs. For performance,
> striping offers the best value. But it also offers no
> protection for your data, so is only desirable when
> data safety is not mandatory. Otherwise, RAID 5 is the
> best combination of safety and performance.
>
> Pat
>