Speed of Raptors in RAID-1

[Magnusfarce]

I'm interested in using two WD 74gb Raptors in RAID-1 configuration on a
higher-end machine. How much speed loss (writing only, I guess) will I see
in RAID-1 compared to a single Raptor by itself?

- Magnusfarce

 

[Jim]

Not as much as some would lead you to believe.

Remember, with RAID1 (mirroring), the HDs are written in *parallel*. And
most writes are buffered anyway, so the *perceived* impact can be difficult
to measure/detect. Windows itself buffers writes too, which is why analysis
tools like SiSoftware Sandra typically disable all software buffering. SATA
improves the situation over PATA as well, since it was possible w/ the PATA
controller for both HDs to be using the same controller, but only one could
be addressed at a time. With SATA, each HD has its own controller.
Finally, don't forget that you usually see improved READ performance w/
RAID1 since you now have TWO HDs that can be read in parallel! IOW, you
lose some, you gain some too.

All in all, as long as the RAID controller and software is good, i.e.,
inherently efficient design, the impact of mirroring is negligible and
hardly perceptible under real world conditions.

If you want the ultimate setup, then you might want to consider doing what I
did. I installed two HDs as RAID0 (stripping), but only for installation of
the OS (expendable). I then installed two other HDs as RAID1 (mirrored),
but only for DATA (NOT expendable). A stripped array is always more
vulnerable than a standalone HD since it increases the opportunity for
failure (lose one and you've effectively lost both). But since it's only
the OS, I don't really care all that much. Again, it's expendable because
it's recoverable from a bootable CD. What I want from the OS is SPEED! On
the other hand, what I want from the mirrored drive is SAFETY! Afterall,
my DATA can NOT be recovered like the OS can. It's difficult to achieve
BOTH speed and safety from the same solution. This is the mistake most ppl
make. They try to accomplish both w/ the same solution. But it's the wrong
approach. Your PC contains two different types of data -- expendable and
non-expendable. Each, ideally, should be treated differently.

But if I only had the capability or finances to support one solution, the
RAID1 solution containing both OS and DATA is the preferred choice. Again,
where talking marginal performance differences here for the typical user,
it's not like it's gonna take TWICE as long or something. Maybe a 5% hit,
if that, under the worst of conditions.

JMTC

Jim

 

[Magnusfarce]

Thanks, Jim, for all the help. This will be the first time I've set up a
systen with the OS and data files on separate drives. My assumption is that
I would keep XP and the main programs (Office, Acrobat, Corel, PhotoShop,
etc.) on the main drive and use the larger secondary drive(s) for the data
files generated by these apps. This would allow the applications (as well
as the OS) to operate very quickly. However, this sounds different from
what you described for your own system. Can you give some advice on this
subject and/or direct me to some information on line? (I don't even know
what to search for!)

- Magnusfarce

 

[Marc]

Not as much as some would lead you to believe.

Remember, with RAID1 (mirroring), the HDs are written in *parallel*. And
most writes are buffered anyway, so the *perceived* impact can be difficult
to measure/detect. Windows itself buffers writes too, which is why analysis
tools like SiSoftware Sandra typically disable all software buffering. SATA
improves the situation over PATA as well, since it was possible w/ the PATA
controller for both HDs to be using the same controller, but only one could
be addressed at a time. With SATA, each HD has its own controller.
Finally, don't forget that you usually see improved READ performance w/
RAID1 since you now have TWO HDs that can be read in parallel! IOW, you
lose some, you gain some too.

As you say, RAID 1 is mirroring. If you hit a wall and look at yourself
in a mirror while doing that, the wall doesn't get more damage.
Just like that, RAID 1 disks don't read any faster, because all they do
is just read the same thing.

Marc

 

[Jim]

As you say, RAID 1 is mirroring. If you hit a wall and look at yourself
in a mirror while doing that, the wall doesn't get more damage.
Just like that, RAID 1 disks don't read any faster, because all they do
is just read the same thing.

You can't really be sure. Afterall, the RAID controller is managing the
read process. What's to say it hasn't been optimized to read differ sectors
across the two HD's, then reassemble them before presentation to the IO
buffers?! Yeah, this may be overly optimistic, esp. with a consumer level
RAID product, BUT, it is possible. There are twice as many read heads
AVAILABLE for the same data, so such an optimization *is* possible, maybe
not *likely*, but possible. Problem is, the vendor never provides proof one
way or the other. Best you can do is runs some tests, such as Sandra is see
what happens.

Jim

 

[DevilsPGD]

In message <[email protected]> Marc

As you say, RAID 1 is mirroring. If you hit a wall and look at yourself
in a mirror while doing that, the wall doesn't get more damage.
Just like that, RAID 1 disks don't read any faster, because all they do
is just read the same thing.

In theory, RAID-1 can be faster -- If the controller supported NCQ (from
a software point of view), and passed read requests to the different
drives simultaneously, then under ideal conditions seek times would be
roughly half of a single drive, or under load, raw throughput would be
double.

However, as far as I know most controllers aren't that smart.

 

[Magnusfarce]

Maybe I should interject: the motherboard I'm using is Abit AN8 Fatal1ty
SLI. I don't know what kind of controllers are on this board.

- Magnusfarce (OP)

 

[Marc]

In theory, RAID-1 can be faster -- If the controller supported NCQ (from
a software point of view), and passed read requests to the different
drives simultaneously, then under ideal conditions seek times would be
roughly half of a single drive, or under load, raw throughput would be
double.

However, as far as I know most controllers aren't that smart.

That could be, but when that happened, the entire idea of RAID-1 would
be changed: Two disks doing exactly the same, so they can go on when one
fails. When the controller has sent a lot of read actions to hard disk
one, and it fails, that could change the stability of a program, and it
might crash. And so, RAID-1 would become a bit useless.

Apart from that: there is one head on the left of the drive, and the
other one on the right, and there is a write action planned on the left,
one of them has to move first, and then even the write actions will be
changed by this idea.

Marc

 

[DevilsPGD]

In message <[email protected]> Marc

That could be, but when that happened, the entire idea of RAID-1 would
be changed: Two disks doing exactly the same, so they can go on when one
fails. When the controller has sent a lot of read actions to hard disk
one, and it fails, that could change the stability of a program, and it
might crash. And so, RAID-1 would become a bit useless.

First, the controller would need to track read requests, so if one drive
failed it could pass the read request to the other drive and go on with
life.

Second, many RAID controllers can't handle a drive failing while the
array is up, you need to reboot before things recover. Not all RAID
controllers are hotswappable in any sense of the word -- These
controllers are still useful, especially in a desktop environment, since
you just reboot and pickup where you left off (with only a single drive,
but without any more data loss then an unexpected reboot)

But the performance increase could be rather substantial, especially for
a drive that mainly reads. This is likely the case for the vast
majority of desktop users, discounting temp files.

Apart from that: there is one head on the left of the drive, and the
other one on the right, and there is a write action planned on the left,
one of them has to move first, and then even the write actions will be
changed by this idea.

Sure, but that's where caching comes into play. Even so, statistically
the average seek time before a write wouldn't be any worse then it is if
all drives are run in parallel.

 

[John Weiss]

There will likely be no perceptible slowdown. If you need benchmarking
software to detect it, it just doesn't matter!
--

 

[digisol]

You will see a bit of speed, I have three raid systems, an old P-
1.4G dual Tualtin striping, a 2600 and a 2800 also striping, it i
better to use a decent PCI raid card rather than the std onboar
version (usually Lite

The raid 1 AMD 2800 Barton will do the same as either of my 3200'
although it depends on what your doing, counting nano seconds can b
pretty hard

Sure Raid is faster but not enough to jump around and have a part
over, just remember if one drive has any problems you lose the lot i
striping

Given the chance again I would buy a good single drive and leave i
there, sure is a heap simpler at times, and cheaper

If your board will take a single SATA drive, that should do the jo
fine, but any decent drive WD JB with 8Mb cache won't be far behind