SATA - Drive size performance question

[John Kimble]

Howdy all,

I'm setting up a RAID 0 array in a new computer I'm
building.

I've narrowed it down to 2 Western Digital Raptors, on
the thinking that since RAID 0 with 2 drives, you have
double the chance to lose all your data, but Western
Digital drives are at LEAST 2x better and more reliable
than any drive out there. Hoping the positive will
balance out the negative.

The question is this, Will the 74 gig Raptor outperform
the 36 gig Raptor? Western Digital has it on their site
that the 74 gig SATA Raptor is outperforming SCSI 10k and
15k drives. Is this the same for a 36 gig drive?

My only other concern is disk space. I definately won't
use all of a 148 gig array, but I may be lacking space
with a 72 gig array. I'm an avid gamer and later this
year will be getting into picture and video editing with
the coming of a new digital camera.

Any feedback is appreciated.

 

[John Kimble]

Forgot to mention that 10k RPM about the Raptors...That's
amazing. Cuts the average seek time of most drives in
half!

That's the kind of performance I'm looking for.

 

[Tom]

Forgot to mention that 10k RPM about the Raptors...That's
amazing. Cuts the average seek time of most drives in
half!

That's the kind of performance I'm looking for.

I wouldn't pay that much for a drive, of which the only performance gain is
the seek time speed. You may be thinking that you are almost cutting that
speed in half, which is true. But when you figure the seek time between a
160gig 7400RPM SATA drive @ avg Seek time 8.5ms to 4.5ms for a 10,000RPM
SATA 74gig (less than half the drive space) for an avg of $70 more, is not
worth it. This due to the fact that most people won't be able to discern 8
milliseconds from 4 milliseconds (thousandths of a second) in seek time
speed, and that one's processor can only process as fast as it can go.

Also, if your going to have two HDDs, I wouldn't recommend a RAID0 array, as
you stated, if it fails, it all fails. Why not a RAID1, then you can fall
back on the same setup, and backup constantly. But IMHO, I would simply set
it up as two separate (basic) drives, and load all of my games and files to
the secondary drive. You figure 74gigs would hold (easily) an avg. of 35
games. You would get bored or outplay the usefulness of most of them before
you would feel the need to keep them stored. You can just as easily play
them from there, as the transfer rates work the same regardless of the
setup.

 

[Guest]

No, there isn't much of a difference in 8.5 ms compared
to 4.5. If it were only one seek I would agree with you.
But there are LOTS of seeks. Those ms's add up.

Plus, I'm not only paying for the extra speed, but also a
5 year warranty and one damn solid drive. My experiences
with Western Digital have been outstanding to this point,
and I have no reason to believe they won't continue to be.

 

[Chen]

You can put a fan that "blows on the harddisks". That should double the
lifetime .

I've been using 2x IBM 180 GXP and 2x 300 gb SATA disks (don't remember the
manufacter) with RAID0 in over half a year now (one year for the IBM disks),
and they haven't failed yet. ;-)

// Chen

 

[John R Weiss]

The question is this, Will the 74 gig Raptor outperform
the 36 gig Raptor? Western Digital has it on their site
that the 74 gig SATA Raptor is outperforming SCSI 10k and
15k drives. Is this the same for a 36 gig drive?

The Raptor 74 is not just a 36 with 2 platters -- it was rebuilt as a new
HD. All reviews I've read say the 74 significantly outperforms the 36.

BTW, I like my Raptor 74s in RAID 0!

 

[Nicholas T. Pappas]

Several weeks ago I also purchased two 75 gig WD Raptors
for a home-brew machine with a motherboard that supports
up to 4 SATA drives for RAID 0 as well. So it appears as
though we are both building machines with similar disk
hardware. (In fact, my case is special also, as it has a
drive cage with a backplane for up to 4 SATA drives; the
backplane allows these drives to be hot-swapped out from
the front, and the cage has forced airflow to keep the
drives cool; more on all that in a minute.)

Before I start, I will tell you that I am a systems
engineer specializing in performance (load and stress)
testing of commercial grade servers. In other words, I
evaluate computer speeds for a living, and I have been
doing this for over 20 years.

Now, regarding your first question about the comparative
speed of the 74 gig vs. 36 gig.: The 74 gig has a lower
seek time so the larger the drive IS faster. So how much
importance does seek time have? Contrary to some other
contributors to your post, I insist it has a lot of impact
most of the time. In the most conservative case, let's
assume my hard-drive is completely defragmented, meaning
each file's allocated blocks are contiguous; and let's
also assume that I am running so much memory that the
entire disk directory structure has been read into RAM and
sits there the whole time (meaning the disk I/O is NOT
constrained due to indexing). For the first example, let's
assume I am copying one large file from one drive to
another PHYSICAL (not logical) drive. This is the only
case where you will NOT see any real improvement, but it
is also a very special case. So here, I will agree with
your prior contributors.

However, as example number 2, consider running the same
copy to another place on the same hard drive - now the
heads are jumping back and forth from the allocated area
for one file to the allocated area of the second file. In
this case, the limiting factor will most likely be seek
time - halve the seek time and double the transfer rate.
For example #3, here is a real life case: when I run my
anti-virus scanner against the entire drive, the process
is now much, MUCH faster with my Raptors than before - as
there are many (10 to 20) "seeks" per second as the disk
is constantly reading in dozens of different files every
second, and each new file means the heads have to go
elsewhere on the drive (even if the ENTIRE drive is
completely defragmented - which it rarely is). Bottom line
is that for most disk-intensive applications (like running
a virus scanner, file I/O at boot time, running back-ups,
etc.) your throughput will double when the access time is
halved.

Before I comment on using RAID, I want to remind you that
drives also have a data bandwidth rating - the maximum
amount of data they can read/write per second. The value
would most likely be the limiting factor in my first
example - copying a large, file from one completely
defragmented physical drive to another completely
defragmented physical drive. For SATA it is 150 MBS vs.
133 MBS for the fastest parallel ATA. Here, seek time has
nothing to do with it, and the limiting factor would most
likely be this value, so the WD 74 gig Raptor would not
even have any significant improvement a sluggish 133 ATA
5600 RPM "home computer" drive such as a Maxtor 250. To
obtain a speed increase when data rate is the limiting
factor, I would have to graduate to the latest SCSI which
is 320 MBS. (And, by the way, the uncontested fastest
drive in the world for several years remains the Seagate
Cheetah with a seek time around 3.7 ms and a data transfer
rate of 320 MBS.) Anyway, my point is that data rate can
also limit overall throughput, but often that is NOT the
limiting factor.

About RAID: RAID 0 is "striping" without redundancy. You
DO NOT have halve your capacity - each file is SPLIT over
two drives, but you still get the same effective overall
storage size. RAID 0 is used strictly for performance:
with a file being split over two drives, you effectively
have twice as many independent heads doing the seeking, so
EFFECTIVE seek time is cut in half, which means around 2.4
ms for a pair of 74 gig Raptors. It is also true that your
chances of losing data due to drive failure is doubled.
(It is the same story with small aircraft: a two engine
prop plane is twice as likely to crash since there are two
engines to fail, and it is too difficult for all but the
best pilots to fly on only one engine with the other one
dragging its prop through the air.) But if your drives are
new AND you are keeping them cool (AND have a good power
supply with well regulated voltage) you should not have to
worry for at least a year or more. Have you owned
computers for a while (or, do you have friends who do? Ask
yourself: whose hard drive has burned out recently? Has
anyone's?) But I also do want to point out that 10,000
RPM drives are HOTTER, and the expansion/contraction from
heating/cooling (by turning the power on and off over the
course of many months) is the single biggest cause of
failure for almost all computer components that are inside
the computer case.) So keep them cool!

Did you ever hear that sound is only as good as the
weakest component in a stereo system? (BTW, usually that
is the speakers.) In the end, for a computer, speed is
always as slow as the slowest component in a chain. And,
for many computer users, it is usually the hard drives.
Now, if it is the hard drive, then its speed (the hard
drive's) is also based on the slowest-component which, for
your Raptor, is probably the data bandwidth of 150 MBS for
large files, and it is the seek time itself for processing
many small files.

Therefore, with all due respect to the other contributors,
they are wrong when they say seek time has little impact.
However, as an almost final comment, I want to add that
the nature of an application will ultimately determine
which computer components are going to actually be the
bottlenecks. The games that I am familiar with are video
intensive, and thus it is the video card that limits their
speed, not the hard drive, and not the CPU. Similarly, for
CAD/CAM and multimedia (such as video editing) it is the
CPU that usually limits overall speed, and NOT the hard
drives and NOT the graphics card! So, in the end, your
question may be moot, at least for games.

Besides, Western Digital would not be able to sell its
Raptor in enough quantity to make it a commercially viable
line if the price increase could not be justified by the
performance gain. But the usually customer is not the home
gamer, it is the enterprise server administrator. These
guys run constant tests to identify bottlenecks - Is it
the network firewall? RAM? CPU? Disk I/O? Perhaps the
encryption/decryption scheme? Is the back-end SQL server
experiencing heavy locking contentions?

Here is what I did, by the way, with my new home PC: I did
try RAID 0 and it was faster. But I wanted to partition
the hard drives for multi-booting (I want to run Windows
XP in both the 32-bit as well as the 64-bit
implementations, as well as Windows 2003 Advanced Server;
and several incarnations of Linux workstation as well as
Linux Server). Hard drive partitioning is not readily
feasible with hardware-based RAID, so I achieved something
like striping by MANUALLY relocating the SWAP FILE to the
beginning of cylinder 1 on the second physical drive and
forcing all my executables to reside on the first drive as
close to cylinder 1 as possible. Now, when I run large
applications or just many small applications at once,
memory fills up and a new program that needs room forces
the paging algorithm to swap out memory pages as it sucks
in new executable code segments, and thus one set of heads
is swapping in while the other is swapping out - just like
striping might due under RAID 0! It may not be quite as
fast as RAID 0, but it sure is damn close. And let me tell
you: those Raptors? They are SCREAMERS! Those other guys
who wrote here? They were just jealous.

 

[John R Weiss]

Before I start, I will tell you that I am a systems
engineer specializing in performance (load and stress)
testing of commercial grade servers. In other words, I
evaluate computer speeds for a living, and I have been
doing this for over 20 years.

Good credentials, but credentials aren't the whole picture when dealing with
analyses...

Before I comment on using RAID, I want to remind you that
drives also have a data bandwidth rating - the maximum
amount of data they can read/write per second. The value
would most likely be the limiting factor in my first
example - copying a large, file from one completely
defragmented physical drive to another completely
defragmented physical drive. For SATA it is 150 MBS vs.
133 MBS for the fastest parallel ATA. Here, seek time has
nothing to do with it, and the limiting factor would most
likely be this value, so the WD 74 gig Raptor would not
even have any significant improvement a sluggish 133 ATA
5600 RPM "home computer" drive such as a Maxtor 250.

The "data bandwidth rating" you note above is NOT a function of the drive,
but of the bus/controller on which it is hosted. Also, that is a maximum
burst rate, NOT a sustained rate. Note that WD terms it the max "buffer to
host" data transfer rate, so it will only approach that rate when the disk
buffer contains the requested data.

That Maxtor (5400, not 5600, BTW) you cite has a max data transfer rate per
drive of 46 MBps. If you read an older Cheetah data sheet, you will find
the max internal ("buffer to disk") transfer rate is 682 Mbps or 63.2 MBps
(gen 6, SCSI 160, 10K RPM version); the latest gen 4, 15K RPM (contemporary
of gen 7, 10K) is 96 MBPS. By comparison, the Raptor 74 weighs in at 102
MBps. [Note: All are mfgr's specs, but should be good for comparison
purposes.]

So, while the external bus may have a max bandwidth as you note, the drive
itself has internal limitations that may be significantly more restrictive
than the external bus. Note that the Barracuda 5400 could not approach the
max theoretical ATA 100/133 bus bandwidth -- even a pair of them with RAID 0
striping -- but a pair of Raptor 74s or Cheetah 15Ks could do so in theory.

Also, if the ATA controller is on the 32-bit PCI bus (via controller card),
it will share the 133 MBps max bandwidth of the PCI bus with everything else
on the bus.

To
obtain a speed increase when data rate is the limiting
factor, I would have to graduate to the latest SCSI which
is 320 MBS. (And, by the way, the uncontested fastest
drive in the world for several years remains the Seagate
Cheetah with a seek time around 3.7 ms and a data transfer
rate of 320 MBS.)

As I note above, 320 MBps is the SCSI bus max data transfer rate, and the
Cheetah itself tops out at 96 MBps. You can only begin to approach 320 MBps
data transfer rate in a 3- or 4-drive array, and then only if all other
components support it.

It is also true that your
chances of losing data due to drive failure is doubled.
(It is the same story with small aircraft: a two engine
prop plane is twice as likely to crash since there are two
engines to fail, and it is too difficult for all but the
best pilots to fly on only one engine with the other one
dragging its prop through the air.)

BAD analogy, and indicative of your lack of true analysis!

Your "too difficult..." statement has a modicum of truth, but ONLY in
limited circumstances -- when the engine fails just after takeoff, when the
airplane's speed is low, the drag is high, and margin for error is minimal.
Your generalization also does not take into consideration airplane
configuration (what if the engines are mounted fore and aft, like the
C336?), jet-powered airplanes, and installed technologies such as
auto-feather systems.

Also, an engine failure in either a single- or twin-engine airplane does NOT
imply or guarantee an impending "crash." MANY airplanes of both types land
successfully with failed engines!


Finally, MTBF rates among various HD brands and types can vary by a factor
of 10 or 100 or more. Even if the risk of data loss is "doubled" by using
RAID 0, you can more than compensate by choice of a more reliable drive in
the beginning.