PC Hardware

Percival P. Cassidy · Guest

It's not just the interface:

http://www.seagate.com/content/docs/pdf/whitepaper/D2c_More_than_Interface_ATA_vs_SCSI_042003.pdf

Perce

Dorothy Bradbury · Guest

You'll get toasted for posting a whitepaper here :-)

SCSI v (S)ATA comes down to usage environment,
plus a high degree of overlap re economics/performance.

(S)ATA HD characteristics:
o Low cost per GB -- several hundred GB per drive very cheaply
o Low cost of performance -- 10k-rpm Raptors high SDTR & low latency
o Eventually multi-user tricks -- NCQ/TCQ in HD now, card/OS later
o Low thermal dissipation -- re server/coloco density & cooling

SCSI HD characteristics:
o High cost per GB -- price premium over (S)ATA
o High cost of highest performance -- 15.3k-rpm drives high SDTR & lowest latency
o Multi-user tricks here now -- NCQ/TCQ in both HD, Card & OS
o High thermal dissipation -- 15.3k-rpm drives need some cooling

As to which is best, it comes down to usage:
o Single User environment is characterised by more SDTR, less access
---- admittedly windows swap/MS-IE/temp do their best, but still largely true
---- thus absolute SDTR is the over-riding factor, not real-world access times
o Multi User environment is characterised by less SDTR, more access
---- multiple users accessing multiple files simultaneously
---- thus absolute seek time, termed IOPS, I/O operations per sec matters

SCSI drives have higher IOPS than (S)ATA drives.
That is due to 1) arguably mechanism re seek times, 2) rotational rpm latency
as a 15.3k-rpm drive beats a 10k-rpm drive, 3) from NCQ/TCQ re-ordering the
seeks to minimise the (slow) electromechanical latency key to IOPS/Multi-User.

I think StorageReview a while back illustrated it:
o On IOPS even quite old SCSI Quantum Atlas outperformed even WD Raptor
o Conversely, on SDTR, even a slowish (S)ATA outperformed 10k-rpm SCSI 'Cuda

That said, SCSI does face competition from (S)ATA.

Cheap (S)ATA created new applications
o Data-warehousing & data-mining are key buyers of (S)ATA
---- often bulk-data mining is not IOPS-critical
---- instead it "customer knowledge = space = cost" critical
o Cheap (S)ATA created backup & recovery latency
---- (S)ATA also provides a solution disk-to-disk-to-tape capability
-------- sure, backup is across a backup network, but N-TB take N-time
-------- more critically restoration of that can cost N-customers

Cheap (S)ATA is quite capable in multi-user:
o (S)ATA may have inferior IOPS, but it offers 1) Substition 2) Cheap
o SCSI is a cost which must compete for a budget like any other h/w
---- SCSI in itself offers a local benefit based on data-set calculations
---- cost-wise, it may offer less vs a 64-bit m/b populated with 32GB RAM
o (S)ATA can offer competent RAID solutions
---- multiple 3ware cards can offer high availability solutions
---- those cards plug into the same motherboards that SCSI do

The trends will be determinant here - and one is clear:
o (S)ATA has opened the door to very high capacities
---- in turn naturally spawning backup solutions using same
o Ever more capacity, means large scale management
---- not everyone can pay for an EMC solution

So I think Smart Software will play a greater role:
o IBMs Ice Cube was not about selling water chilling plant
---- it was about reducing data-warehouse staff mgt costs
---- plus at the same time boosting availability/reliability
o Smart s/w could negate differences between (S)ATA & SCSI further
---- particularly re the perceived end-user/application latency
---- s/w costs nothing to reproduce across many buyers

Eventually (S)ATA will offer NCQ/TCQ through to the O/S, and so
one of SCSI drives components of low IOPS will be removed. So it is
likely SCSI will move to higher rpm, I believe 18,000rpm & 21,000rpm.

Hard drive form-factor is not static:
o HDs are moving to 2.5" form-factor from 3.5" -- /eventually/
---- 2.5"-ATA are limited to 7,200rpm (SATA soon to 5400rpm)
---- 2.5"-SCSI are already at 10,000rpm (Seagate Savvio) at a (high) price
o A 2.5" HD size is of interest to Coloco/Machine-Rooms
---- airflow - 1U will be tough to cool with dual-core dual-CPUs & 3.5"
-------- however you can fit 4x 2.5" in the place of 1x 3.5"
-------- so 56dB(A) per 50x40x38mm fans may not be needed quite yet
---- availability - 1U takes 3x 3.5" - just enough for RAID-5
-------- conceivably 1U can take 12x 2.5" - enough for higher RAIDs
---- wattage - 2.5" have lower dissipation than 3.5" counterparts
-------- 15.3k-rpm SCSI are already a 2.5" platter however
o If only 2.5" will fit & you want the highest performance, it's SCSI
---- only Seagate Savvio offer 10,000rpm in a 2.5" form-factor

The key to SCSI, beyond NCQ/TCQ, will remain rotational latency.
Somewhat like Mini-Computers, the PC of (S)ATA is nibbling away.

I've not mentioned mechanical/warranty aspects of SCSI:
o Arguably the drives could be the same
---- failure of your larger installed (S)ATA based wouldn't go down well
o Warranty in itself isn't indicative
---- Seagate recently put a 5yr warranty on their non-SCSI drives (not all IIRC)

So which is best comes down to the application:
o Is the data-set so IOPS critical that there is a business cost to not using SCSI?
---- or is the money better spent elsewhere, eg, 64-bit Xeon & buckets of RAM
o Thus I expect financial services e-commerce/transaction servers to remain SCSI
---- whereas multi-user servers using (S)ATA may nibble up the hierarchy

Might come down to whether the economic buyer can do the calculations to prove
if on performance the SCSI solution pays it way re opportunity cost of capital. Some
companies probably keep reliability data on their (S)ATA & SCSI installations, so it
may be there is a difference there and that is factored in to the cost-benefit analysis.

SCSI does seem to be the "Mini-Computer" w.r.t. premium vs Moores law.
Comes down to where the bottleneck is, and budgetary constraints/need.

I am surprised we haven't seen higher RPM SCSI drives.
o End of decade will see 2.5" drives into the hundreds of GB, maybe 800GB
o That could create some interesting solutions - and large e-waste-paper :-)

FEMA to National Identify Cards database demands aren't going to shrink, along
with future computerised health care records, right down to insurer/finance records.
IBM will also be pushing for integration of all information - employment, CVs etc.
Then there will be data-mining linkage to the tax computers, since it seems we have
a recovery which is light on taxes, but heavy on spending - as California has found :)
Roadside pay per mile, pay to park, speed & time monitoring also integrated.

That lot, plus data-mining in business smells of capacity being key.

Of course, will it be easier to manage that - or an XP upgrade as DWP UK showed...
The sell-side rarely asks 'should we', rather than 'could we' :-)
--
Dorothy Bradbury

Guest · Posted: Thu Dec 02, 2004 8:02 am Post subject: Re: Why SCSI rather than ATA?

On Tue, 30 Nov 2004 18:15:08 UTC, "Percival P. Cassidy" <Nobody@NotMyISP.net> wrote:

| It's not just the interface:
|
| http://www.seagate.com/content/docs/pdf/whitepaper/D2c_More_than_Interface_ATA_vs_SCSI_042003.pdf
|
| Perce

Because of neat things like intelligent co-processed scsi cards?
--

Percival P. Cassidy · Guest

I don't recall that they even mentioned things like that. A couple of
things that struck me were the fact that SCSI drives are designed to be
used in multi-drive setups and therefore for less mechanical interaction
and better cooling. ATA drives, OTOH, are expected to be one per box, so
tend not to take things like that into account.

Perce

On 12/01/04 10:02 pm drgynfly@_.mv.com tossed the following ingredients
into the ever-growing pot of cybersoup:

Guest · Posted: Thu Dec 02, 2004 10:07 am Post subject: Re: Why SCSI rather than ATA?

On Thu, 2 Dec 2004 03:37:00 UTC, "Percival P. Cassidy" <Nobody@NotMyISP.net> wrote:

| I don't recall that they even mentioned things like that. A couple of
| things that struck me were the fact that SCSI drives are designed to be
| used in multi-drive setups and therefore for less mechanical interaction
| and better cooling. ATA drives, OTOH, are expected to be one per box, so
| tend not to take things like that into account.
|
| Perce
|

The fastest drives I keep around are Quantum 10k u160, which seem
to run relatively cool to the touch. I really should modernize, though
I've been planning to switch platforms for a while now.

On the other hand, I've tested seagate 7.2k drives where the drive's
heatsink was nearly hot enough to be a fire hazard.

I think co-pro scsi cards are excellent. Is there anything comparable
for ATA?

|
| On 12/01/04 10:02 pm drgynfly@_.mv.com tossed the following ingredients
| into the ever-growing pot of cybersoup:
|
| > On Tue, 30 Nov 2004 18:15:08 UTC, "Percival P. Cassidy" <Nobody@NotMyISP.net> wrote:
| >
| > | It's not just the interface:
| > |
| > | http://www.seagate.com/content/docs/pdf/whitepaper/D2c_More_than_Interface_ATA_vs_SCSI_042003.pdf
|
| > Because of neat things like intelligent co-processed scsi cards?

--

R Weis · Guest

drgynfly@_.mv.com wrote in message news:<jPvZ0GeBvu6y-pn2-dRveWvEQZPou@localhost>...

Dorothy Bradbury · Guest

	PC Hardware Forum Index -> SCSI	All times are GMT
Page 1 of 1