I thought this was informative and interesting considering there have been a number of threads debating the use of RAID0 over a single drive. But, when you consider the intent and purpose of RAID in general, it's not a real surprise that a cheap RAID0 array can outpace a fast single drive.
I remember several threads here where many supposedly knowledgeable posters told me I was wrong when I said there's a noticeable performance gain on a desktop PC by using raid 0. Well I guess this article massively proves they're wrong.
However, the real point is missed in the article. I doubt if hardly anyone is really making a buying choice between a single raptor and two other hard drives as if you need tons of space you wouldn't even consider a single raptor in the first place.
The real point that they missed in the article is how good the case is for putting 2 raptors in raid 0, that way you get both large and crazy fast storage.
I remember several threads here where many supposedly knowledgeable posters told me I was wrong when I said there's a noticeable performance gain on a desktop PC by using raid 0. Well I guess this article massively proves they're wrong.
RAID is a very context dependent technology which is entirely dependent on the RAID controller used. You see a much more significant increase in performance with a high end controller vs. onboard on an enthusiast motherboard. RAID does a very good job in proper applications, but certain instances it doesn't do jack. Something to keep in mind is that several of these benchmarks were synthetic and won't mimic what you see in an average day by an average user.
People who use large files, have little RAM (ergo use the page file constantly), work with scratch disks in adobe, etc. will see noticable improvements with RAID 0. In gaming load times, day to day access of the internet, M$ office files, etc. will see little improvement (if any).
As for the article.... Anyone notice the Raptor's write speed being capped at 150MB/s in the first set of benchies? Doesn't that strike yall as odd? I realize the interface of the Raptor 150 is SATA 150 and therefore limited to 150MB/s, but SATA is a point to point connection. So if you are using 2 SATA connection you have a theoretical bandwidth of 300MB/s (150MB/s for each drive). Unless, going through the RAID controller merges these two lanes and is then capped by the resulting SATA 150 spec...
What I find interesting is that a Raid 0 setup with Raptors is about the only setup that is fast enough to warrant SATA2-300, and yet they're about the only drive that's still using SATA-150. D'oh!
Well I guess this article massively proves they're wrong.
The only thing this article proves is how the quality of reporting on THG has declined through the years. Of course a RAID array is going to trounce a single drive in STR (sustained transfer rate) and related benchmarks. For desktop use, its still going to be slower, however. The article recycles the same old misconceptions about RAID, its intent and purposes, and its performance implications.
The first hint the author is clueless is his statement that, "Though a RAID 0 array doesn't reduce access time." A Raid array doesn't reduce access time true...but it doesn't leave it untouched either. It increases it. A single drive has mean rotational latency of 1/2(1/rpm)...whereas for a three drive array, the figure is 0.875(1/rpm) for writing and 0.75(1/rpm) for reading.
RAID 0 massively boosts STR, but it increases rotational latency as well. For server usage patterns, STR tends to trump latency...especially on wide arrays as STR increases linearly, but latency increases logarithmically. But for desktop usage patterns and narrow arrays, the situation is different. In nearly every possible situation, you'll get better performance out of a two-spindle independent disk setup, rather than by linking those two drives into a RAID 0 array.
Another hint the author doesn't fully understand RAID is when he compares a "fast" RAID 0 array to "slow" RAID 1. For writing, this is invariably true. However, for reading small-to-medium sized files, RAID 1 with a savvy controller can exceed the performance of RAID 0. The reasons why lie in rotational latency again. RAID 0 must meet the expectation value of all drives in the array, whereas RAID 1 can take the first drive able to answer the request.
I remember several threads here where many supposedly knowledgeable posters told me I was wrong when I said there's a noticeable performance gain on a desktop PC by using raid 0. Well I guess this article massively proves they're wrong.
I think the conclusion in the article is a little off base, as is your statement. The article does show that RAID has noticeable results, especially when manipulating large files. 100+MB/sec reads and writes is amazing. If you do large file manipulation, or work with large files in the form of videos, music, CAD, etc. etc., the RAID setup will double the speed of transers and halve your wait time. However, if seek time is your biggest concern, like with DB access or booting windows, then spindle speed is your only concern.
Going to a high speed spindle like the Raptor or a high throughput setup like the RAID 0 configuration will lead to noticeable results over a single 7200 RPM drive no matter what, but which is best depends on what you're doing with your computer.
I remember several threads here where many supposedly knowledgeable posters told me I was wrong when I said there's a noticeable performance gain on a desktop PC by using raid 0. Well I guess this article massively proves they're wrong.
I have never owned a Raptor. If they dropped in price by 25%, I would consider pruchasing 2, and I would run them in RAID 0.
The 2 $40 drives I have in RAID 0 absolutely make a huge difference over any other single drive I have ever owned. No doubt about it, the performance is there, and can be felt.
No matter what kind of drive setup you choose to run, backup, backup, backup cannot be stressed enough.
Aside from access time, which can be managed
and minimized with NCQ and regular defragmentation,
the most important limiting factor, overall, is the rate
at which raw data passes under the read/write heads.
Since the binary digits are passing under the
read/write heads at less than 100 MB/second,
even on the Raptor, an interface speed
of 150 MB/second is still more than adequate,
overall.
I believe this is the reason why WD has not
increased the Raptor's interface speed,
even though this product has been on the
market for several years now.
It's only when a file is small enough to fit
entirely within the HDD's cache that
the faster interface speed begins to
make a noticeable difference. Seagate's
website has a few documents which explain
this phenomenon in detail.
On the margin, a 16MB cache should be
mandatory in any SATA-2 HDD that also
has a 300 MB/second interface. A 32MB
cache would be even better (but not
currently available in a single HDD).
We've come to believe that adding more RAM
and configuring the additional RAM as a
RamDisk or SuperCache is a much
cheaper way to accelerate regular
disk I/O, like browser cache and
swap file access. Just compare
10 ms access times with the speed
at which standard DDR2-6400 oscillates
i.e. nanosecond cycle time and
NO READ/WRITE HEAD MOVEMENT
-AND-
NO ROTATIONAL LATENCY!
And, Windows XP/Pro will still "swap"
a program even if there is unused RAM
available for that program to use.
Why force a "swap" to use a relatively slow HDD
when so much unused RAM is available?
RamDisk Plus configures a region of RAM
to operate like any other letter drive,
without requiring BIOS changes.
It also saves and restores the contents
of each RamDisk between shutdowns
and startups: very slick implementation!
Cheers for that. It's always been a bit of mystery why you can add more RAM and XP responds by using an even bigger swap file. Do you think it would be ok to sacrifice 512k of 2gb to a ramdisk, or are we talking about getting another 2gb and ramdisking that?
As someone who needed to make the same choice a few weeks ago. My requirements were performance and reliability. I wasn't particularly concerned about capacity as I have another drive for those purposes.
I had my system hard disk crash, which resulted me being off the air for almost a week while I got a replacement.
I chose 2 x 400GB WD RE2 drives but I have connected them using RAID 1 mirroring instead of RAID 0 striping.
Sure I lose 400GB of disk space but I gain in read performance, get slightly worse write performance, and I have the best reliability as I can keep on running even with 1 of the disks failing.
Considering I was comparing 1 x 150GB Raptor vs 2 x 400GB RE2 disks, I think the RAID1 solution was the best for me considering my requirements.
Sorry if you felt it was overly harsh, but the old THG never would have released a steaming pile of doggie doo like this. It's bad enough for the article to announce that RAID increases STR and latency like it's the discovery of fire or something. But then to conclude from that-- without a single real-world application benchmark-- that RAID is "faster" is worse than outrageous. It's criminally inept.
There is nothing in this article that couldn't have been deduced from a pencil and paper alone, without testing. Nothing but its conclusions, that is...conclusions that are wholly and completely incorrect.
Sure I lose 400GB of disk space but I gain in read performance, get slightly worse write performance, and I have the best reliability as I can keep on running even with 1 of the disks failing.
I think this will depend on your controller. The cheap onboard ones I doubt will take advantage of that opportunity, but the high end dedicated cards should (at least any worth their money). I don't know what you use as a controller so you could very well be right, but it is worth pointing out.
A Raid array doesn't reduce access time true...but it doesn't leave it untouched either. It increases it. A single drive has mean rotational latency of 1/2(1/rpm)...whereas for a three drive array, the figure is 0.875(1/rpm) for writing and 0.75(1/rpm) for reading.
I don't necessarily disagree. However, I do wonder where you get this formula from.
I understand the first one (that is easy), but where does the 3 drive array formula come from (and the read/write formulas)?
I see what you're driving at, but haven't seen it documented before.
I remember several threads here where many supposedly knowledgeable posters told me I was wrong when I said there's a noticeable performance gain on a desktop PC by using raid 0. Well I guess this article massively proves they're wrong.
However, the real point is missed in the article. I doubt if hardly anyone is really making a buying choice between a single raptor and two other hard drives as if you need tons of space you wouldn't even consider a single raptor in the first place.
The real point that they missed in the article is how good the case is for putting 2 raptors in raid 0, that way you get both large and crazy fast storage.
Like some noted above, the article is a bit misleading (and yes, a little below the standard we have come to expect fro Tom's over the years).
If you notice in the only benchmark that resembles real-life scenario for desktop (as in NOT sequential transfer) single Raptor beats 7200RPM Raid. Furthermore, the single Raptor beats the Raptor RAID in windows start-up time bench. Like you are so rightfully noted, this HAS been talked about, and the point it the real world on desktop does not justify the use of RAID.
Now, if you like running synthetic benches time and time again and getting off on the results, be our guest. Come to think of it, be our guest either way. People that have said something about RAID0 in these forums simply try to explain to people that "your mileage may vary," and quite dramatically, depending on the usage scenario. If you notice the same people that advocate against the use of RAID0 for typical desktop scenario recommend it for people who intend to do a lot of multimedia work, where sequential transfers are utilized quite often. It's hard to recommend RAID0 for servers; you would really like some redundancy there.
Now, going back to the article, it is quite a shame that the focus was on synthetic transfer rate benches. More real life benches would have been great.
