Storage

[Alan Dayley]

Trent Shipley said:
> There has been a discussion about drive interface standards on the
Postgresql
> general list.  I don't understand it.

Your questions are very broad.  Maybe some context of the discussion would
help to narrow the focus a bit.  Let me provide some specific bullet
points.

> What is the difference between IDE, SSCI, ATA, and SATA?

- IDE (Integrated Drive Electronics) is the early marketing term for ATA
(Advanced Technology Attachment).  They are really the same. [1]

- You didn't mention ATAPI (Advanced Technology Attachment Packet
Interface) with started as an extension to ATA for support of removable
media drives like CD-ROM and Iomega ZIP drives.  Probably because ATAPI
hard disks don't exist.  In any case, ATAPI is part of the ATA protocol
specification now.

- SATA (Serial ATA) is an update to ATA to transfer the same command set
over serial channels instead of a parallel bus. [2]

- Because of SATA, the term PATA (Parallel ATA) is now being applied to
the original ATA devices.  ie. IDE == ATA == PATA.

- SCSI (Small Computer System Interface) has been around a long time
compared to ATA with it's roots going back to 1979.  It handles devices of
many types (scanners, tape drives, media changers, etc.) besides hard disk
drives.

- You did not mention a new tech called SAS (Serial Attached SCSI) which
is similar in nature to the ATA-->SATA evolution.  That is, transmitting
the same SCSI commands in serial packets. [4]

> What are the
> implications for:
> drives on notebooks and PCs

This maybe best answered by looking at the goals of each protocol.  I'll
reduced them to ATA vs. SCSI since the rest are flavors of one or the
other.

The original goal of ATA was to provide an inexpensive "consumer grade"
hard drive interface.  SCSI, from the beginning, was intended for reliable
"systems or industrial grade" data operations, including higher
performance.  As a result of these somewhat opposing goals ATA
technologies are common in the consumer computer world and ATA devices
tend to be less expensive per MB.  SCSI devices are very common in
industrial applications, servers and main frames, for example and are more
expensive per MB.  SCSI fans look down on ATA devices as cheap and prone
to failure sooner rather than later.  ATA fans look down on SCSI devices
as overly expensive and over engineered.

Truth is that, whether by side effect of improving technologies or by
design, ATA is evolving to be just as good and fast as SCSI.  Believe it
or not but there are still far more SCSI storage systems in use today than
ATA but ATA use is multiplying very fast and is taking over uses where
SCSI was traditionally deployed.  It used to be that finding ATA drives
with high duty cycles and MTBF was difficult but that is changing.

The bottom line is that you'd be hard pressed to find a notebook with a
SCSI hard drive in it.  Similarly you'd be hard pressed to find a
mainframe or heavy industrial computer with an ATA drive.  But the two are
converging both in terms of price and capabilities.

> low TB drive arrays
> SAN?

Drive arrays and SAN and NAS, etc. are technologies either combining or
sharing (or both) storage to a host or across a network.  This is a layer
"above" the actual hard disk interface layer represented as SCSI or ATA or
one of it's flavors.  In other words, you could have a SAN server that
uses ATA drives and another that uses SCSI and both would interact with
the network layer exactly the same.  Clients using the storage would not
know what hard drive protocol is used.

So, how the hard drive protocol effects the network storage technology
manifests in only in how reliable, fast, etc. the interface operates.  One
who believes SCSI is best in a SAN service probably believes it is best in
every other situation for the same reasons.  Some dismiss ATA as a viable
server interface because of it's history as consumer grade.  I can accept
such a thought as a rule of thumb but not as universally true.

A server is general powered and operating for 24/7 so the most important
specification is the drive's duty cycle.  How many hours per 24 is the
drive designed to be operating?  The duty cycle of most ATA drives is 10
hours or less per 24.  (There are more and more that have a 24 hour cycle
but they'll cost you.)  Using such a drive in a server is not "wrong" if
you have a good data backup process that you actually follow and you are
willing to save money on each drive but probably buy them more often.

Corrections and discussions are invited.

Alan

[1] http://en.wikipedia.org/wiki/Integrated_Drive_Electronics
[2] http://en.wikipedia.org/wiki/Serial_ATA
[3] http://en.wikipedia.org/wiki/SCSI
[4] http://en.wikipedia.org/wiki/Serial_Attached_SCSI

PS Wikipedia is wonderful!