Woohoo! Just ordered up 2 new 74GB 16mb Raptors!

[Brahmzy]

Get 'em now at NewEgg for $139 AMIR.

Gonna RAID0 'em.

Dang, I just realized how much I spent just now. Straight took a big ol' chunk out of the PayPal account. :( I still have Mobo, RAM, Proc & GPU to buy. :(

I hope they're fast!

I had a single 1st gen Raptor 74GB 8mb HDD and I thought it was fast. These should scream.

[troisanh]

Gratz man but why not single 150g raptor? Working great for me coming from raid0 raptors myself. Gota love the quiet. :D

[AMD_Gamer]

nice dude

[osalcido]

Dang, I just realized how much I spent just now.

WD just pooned you. :)
I have a single 8mb raptor and IF I need to upgrade (Did you really need to upgrade from a single raptor ? You must be running a server or something to need that much speed) I'll just buy another to raid0

[dtess17]

Get 'em now at NewEgg for $139 AMIR.

Gonna RAID0 'em.

Dang, I just realized how much I spent just now. Straight took a big ol' chunk out of the PayPal account. :( I still have Mobo, RAM, Proc & GPU to buy. :(

I hope they're fast!

I had a single 1st gen Raptor 74GB 8mb HDD and I thought it was fast. These should scream.

any reason you didn't just buy the 150?

[Brahmzy]

I wanted blistering speed/performance, not capacity so much.

Heck, if the new 16mb 36GBers were out, I prolly would've bought them. I haven't done RAID in about 5 years, so I figure I'd give it another shot. And $60 off the pair seemed to help the decision.

Also, the new 74GB 16mb are faster in some areas than the 150GBers.

I love fast HDDs - if solid state was available, I'd have it. iRAM's only 4GB and almost twice the price of these two new drives I bought.

Hopefully I'll like it. They'll be in a P180b, so hopefully they'll be quiet.

[chaikovski2002]

I thought single raptor 150 or the new 74gig one is faster in single user environment than 2 raid0 ones?

[TeeJayHoward]

I thought single raptor 150 or the new 74gig one is faster in single user environment than 2 raid0 ones?
In most scenarios, that is the case. However, if he were doing something which requires a higher sustained transfer rate, such as raw DV->raw AVI transcoding, RAID-0 may be faster.

Edit: Just noticed you said Single USER, not single drive. Doh! Ignore this post.

[Brahmzy]

http://babelfish.altavista.com/babelfish/trurl_pagecontent?lp=nl_en&url=http%3A%2F%2Ftweakers.net%2Freviews%2F621

2 150GBer are faster in RAID0, but I ain't got $500!

[brachy33]

Congrats! I absolutely enjoy my new 74GB 16MB Raptor. It has noticeably sped up app load times and system boot times, not to mention Diskeeper only takes 15 sec for a full defrag! Gotta like that.

[mdameron]

Would RAID-0 decrease install time? I re-installed Diablo 2 for the first time in ages yesterday, and I remembered with agony the 10 minute install. :D

Though I noticed throughout the actual install, CPU usage was minimal... I want blazing installs!!!

[Chowder Head]

Quick question.

This drive would be still fast as crap just by itself, right? Just a single drive, no pairing.

[drizzt81]

Would RAID-0 decrease install time? I re-installed Diablo 2 for the first time in ages yesterday, and I remembered with agony the 10 minute install. :D

Though I noticed throughout the actual install, CPU usage was minimal... I want blazing installs!!!
depends: are you installing from the install media, i.e. CD-ROM or from a virtual drive? If the former: no. If the latter: maybe.

[drizzt81]

Quick question.

This drive would be still fast as crap just by itself, right? Just a single drive, no pairing.
yes. Depending on whom you're asking and what your workload is, there may be no performance increase from RAID-0-ing the drives.

[Ockie]

Gratz man but why not single 150g raptor? :D


Seconded.

[Jason711]

prepare to be disapointed with raid 0.

[Brahmzy]

And why is that?

[Jason711]

it only helps out in a couple very niche areas. it isnt noticeable at all in day to day use. and it doubles the chance of data loss.

[mdameron]

Who cares, it's a learning experience. I'm gonna pick up 2 80 gig SATAs just to pud around with... As of right now, I don't even know how to set up a RAID 0 array... I damn sure will by the end of the next pay period.

[chaikovski2002]

deleted

[Jason711]

are those the latest 74gb drives?

[mwarps]

meaningless image

Those were the "Old" 74GB raptors.

[gman]

Get 'em now at NewEgg for $139 AMIR.

Damnit, I just got mine last Thurs. and now their frikin rebating them. Man that's pissy.

[gman]

So 2 Raptor 74's, built off the same drive as the 150, in RAID is not faster than one 150?

Sorry guys, I know you'll argue this up and down, but I just don't understand how that can be.

[Poncho]

So 2 Raptor 74's, built off the same drive as the 150, in RAID is not faster than one 150?

Sorry guys, I know you'll argue this up and down, but I just don't understand how that can be.


Some people are just so blinded by their hate of RAID0 that they will not accept the simple reality that given the same drives, RAID0 will be faster. What I don't understand is why people have such animosity towards RAID0 and feel the need to flame people who decide to run THEIR system that way. I would get it if somebody came in and said that their RAID0 system was 50% faster with the RAID than without.... ok, say something then. But when some guy comes in and is all excited that he bought a couple drives to run RAID0 don't just reply with "prepare to be disappointed" :rolleyes:

[chaikovski2002]

Well correct me if I am wrong, but people report raid0 has no real benefit for desktop performance. All it does is earn bragging rights because it scores higher in synthetic benchmarks, so if you like playing hdtach or sandra all day long then knock yourself out. In the real world raid0 helps very little. Also you increase your chances of data loss by 2, thats why people were saying erarlier on this thread to get the 150gig version and a large dump drive for evertything else. I got the raptor150 two months ago and a large dump drive, and I couldnt be happier! Also keep in mind were still on sata150, no current sata drive even saturates the sata150 channels yet.

[Poncho]

All it does is earn bragging rights because it scores higher in synthetic benchmarks, so if you like playing hdtach or sandra all day long then knock yourself out.


So... what you are saying is RAID0 is the AMD of storage solutions. :D Sorry... couldn't resist. But really though, so what. It is, at the very least, the same speed as a single drive for some stuff and faster for others. And bragging rights are nice to have. Do you think I have the equipment in my sig because my old system was slow? I'm also planning on moving to a dual processor box soon, do you think that's because this one is slow? Sometimes.... it's nice to throw the e-penos down on the table. :D

Also you increase your chances of data loss by 2, thats why people were saying erarlier on this thread to get the 150gig version and a large dump drive for evertything else.

This is such BS. Sorry man.... but I've ran RAID0 on numerous systems and never once had a problem. I've had single backup drives fail.... guess what, you lose all the data that way to. People that boot from RAID0 need a backup option just like anybody running a single boot drive. I would NEVER store critical data on any single drive solution. That's what RAID1 and RAID5 is for. BTW... what's to say that he doesn't have a "dump drive" in addition to the RAID array?


Bottom line is that there are no reasons that can be given for the amount of RAID0 bashing in this forum. It's lame.... let people do what they want. If somebody asks whats better.... fine tell them, but flaming people who just made a purchase (could be big or little depending on the income) is fucking lame. Not saying you did this.... just speaking in general.

[ivzk]

I can't be the only one wondering WTF are you talking about when you mention AMD.

[mdameron]

If you read what he quoted, it makes perfect sense. It wasn't meant to be inflammatory, hence the smiley.

[/off topic]

[WhyYouLoveMe]

I can't be the only one wondering WTF are you talking about when you mention AMD.

You're not.

[Poncho]

I can't be the only one wondering WTF are you talking about when you mention AMD.


Faster in benchmarks, no difference (compared to intel) in real life. Sorry... been spending too much time in the Intel forum. :D

[ivzk]

You've must of had that one ready for the past 2 years. ;)

[TeeJayHoward]

Sometimes.... it's nice to throw the e-penos down on the table. :DI agree :D
I would NEVER store critical data on any single drive solution. That's what RAID1 and RAID5 is for. I disagree. The implications made by this statement, rather intendeded or not, is that with a RAID-1 or RAID-5 solution, backups are unnecessary. Running drives in RAID-1 or RAID-5 only prevents a single kind of mishap - a drive failure - from destroying all your data. PSU failures, Lightning strikes, Overheating, Shorts, and User Error can still destroy data on even a RAID setup. Critical data saved on a single drive and backed up off-site is a far superior solution to data stored on a RAID-1 or RAID-5 array. Of course, RAID-1 plus offsite backup is superior to this :)

[WhyYouLoveMe]

PSU failures, Lightning strikes, Overheating, Shorts, and User Error can still destroy data on even a RAID setup. Critical data saved on a single drive and backed up off-site is a far superior solution to data stored on a RAID-1 or RAID-5 array. Of course, RAID-1 plus offsite backup is superior to this :)

But what about nuclear disasters or global destruction? Crap now we need an off-planet solution.

You see where this is going? To each his own.

[Poncho]

I agree :D
I disagree. The implications made by this statement, rather intendeded or not, is that with a RAID-1 or RAID-5 solution, backups are unnecessary. Running drives in RAID-1 or RAID-5 only prevents a single kind of mishap - a drive failure - from destroying all your data. PSU failures, Lightning strikes, Overheating, Shorts, and User Error can still destroy data on even a RAID setup. Critical data saved on a single drive and backed up off-site is a far superior solution to data stored on a RAID-1 or RAID-5 array. Of course, RAID-1 plus offsite backup is superior to this :)


Eh... for most users a RAID5 or RAID1 solution is more than enough redundency. In a business setting I agree... but you don't need triple backups of your MP3s and porn. Personally, I like having my stuff on a large single drive in my main system and that same data on my software RAID5 array. Now the only thing that will kill that array is a 2 drive failure, for whatever reason, and should that go i've got my single drive as backup and vice versa. Now... should my apt. burn down yea, I'll lose my data. Big deal... it'll suck, but not life or death. And if I need to secure any VERY important documents it's all about the safe, but then again.... docs like that shouldn't be kept on a home PC IMHO.

[TeeJayHoward]

But what about nuclear disasters or global destruction? Crap now we need an off-planet solution.

You see where this is going? To each his own.
It all depends on how critical the data is to you. If it's things you'd just hate to lose, store it offline in a closet somewhere. (USB drives FTW!) If your business depends on it, store it off site. If your financial life depends on it, store it offsite. If the world depends on it, store it off-planet. If the galaxy depends on it, store it in another galaxy. If existance itself depends on it, you're screwed :)

Eh... for most users a RAID5 or RAID1 solution is more than enough redundency. In a business setting I agree... but you don't need triple backups of your MP3s and porn. For most users, you're absolutely right. RAID-5 for incredibly large quantities of files (pr0n). RAID-1? Never. Why bother giving yourself an extra chance to lose the data when, for the same cost, you can just unplug the drive and stick it in a closet?

[Jason711]

raid 0 just isnt worth it. if you wanna do it as a learning experiance, or just for fun, thats fine.

[general]

I should point out that RAID-0 doesn't double the chance of data loss. The number one cause of data loss when dealing with RAID-0 is corruption. This occurs more frequently than drive failure. The true chance of data loss then is more than tripled.

Those doing professional data work need RAID-0, but for regular desktop purposes there isn't much point. The drawbacks are the extra cost, extra heat, noise and power. You'll enjoy nice and fast disc performance because you're using a Raptor. The RAID-0 part won't make any difference.

[Poncho]

The drawbacks are the extra cost, extra heat, noise and power. You'll enjoy nice and fast disc performance because you're using a Raptor. The RAID-0 part won't make any difference.

- Cost is purely subjective. 10 bucks is a lot to some people, $1000 isn't a lot to others.
- Heat, the heat from one more drive is negligble.
- Noise, if i can't tell a difference between 1 raptor and 2 in my case which is VERY quiet then I doubt anybody else will. It's not like we are talking about quiet drive, raptors are louder by themselves
- Power, yea those few extra watts are killer. :rolleyes:

Look... I've ran single Raptors and RAID0 raptors and can tell you there is a difference in normal, everyday usage. Is it a HUGE difference? Not at all... but there is a difference.

[Poncho]

Also... it should be noted that the data loss argument is moot since you should NEVER not have your data backed up. RAID0 or single boot drive it doesn't matter... back up your data and it's a non issue.

[Brahmzy]

Holy smokes this thread has gotten out of hand, hehe. Sheesh, if I knew this was going to happen, I prolly wouldn't have posted it. I'm gonna RAID 'em, and if I don't like that, I'll use 'em as single drives, but I'm pretty dang sure I'll love 'em in RAID0. I know for a fact RAID0 makes a difference in real world single user because I've done it. As far as noise and heat, they'll be sitting in a P180b on rubber isolators with a fat 120mm suckin' air over 'em. Check this out if you haven't already seen it:

And this is the OLD SCHOOL 74GB Raptors-not the new beasts. REAL WORLD testing here folks....

http://www.ocforums.com/showthread.php?t=445800

[Jason711]

id like to see something other than a stopwatch being used.

[Poncho]

id like to see something other than a stopwatch being used.

of course you would :rolleyes:

[Brahmzy]

id like to see something other than a stopwatch being used.
Wow. It's all coming together now...

There could be real world timing of double the performance and you'd still swear RAID doesn't make a difference.

Got it.

[TeeJayHoward]

Wow. It's all coming together now...

There could be real world timing of double the performance and you'd still swear RAID doesn't make a difference.

Got it.
Dangerously close to a flame...

I must admit, the link was an interesting read. As the above poster said, use of something more accurate than a stopwatch is desirable, but, given the times mentioned, it most likely wouldn't make a difference.

Congratulations on your new purchase. You should be happy with them, RAID-0 or not. Raptor = ePeen++

[Jason711]

of course you would :rolleyes:

yes, of course i would. :)

[unhappy_mage]

Wow. It's all coming together now...

There could be real world timing of double the performance and you'd still swear RAID doesn't make a difference.

Got it.
He even shows more than twice the speed, and people still won't believe his results! Jeez!

The above remark is sarcasm. If the application is disk-bound, one would expect two disks to be twice as fast, at most. The fact that he gets 264% performance out of two disks versus one in Photoshop CS2 and 268% out of bootup time tells me his testing is as broken as it's possible to be. Also, notice the skimpy details - what was his cooling setup (for the drives, case, CPUs) like? Was a virus scanner running? When is the bootup time test started and stopped? What services were running? Was the machine networked?

Finally, where's the test with the 2*150, 150, 2*74, and 74? The 2*74 may well be faster than one, but if it's slower than a 150, it's a moot point. There'd be no reason to buy a pair of 74s if a 150 is smaller, cooler, the same price, *and* faster.

http://www.hardfolding.com/ftag1.php/mem/150072.png (http://www.hardfolding.com?go=38&tm=33&id=150072)http://www.hardfolding.com/utag1.php/mem/428/1.png (http://www.hardfolding.com?go=36&id=428&type=1)

[Brahmzy]

What are you arguing? That RAID 0 makes no difference whatsoever?

I know the gains may be small in some areas, but I don't care. Don't dump on RAID because you (with your particular budget) don't think it's "worth it." If my PC boots up onto the desktop 3 seconds faster, it's worth it to me. If BF2 takes 5 seconds less to load a map, it's worth it to me. Nuff said.

Another one below...pay attention to the 74GB model numbers in some of the benchies. Some are the old and some are the new. Oh wait, these are synthetic benchmarks-we can use these! Silly me. Ugh.

http://babelfish.altavista.com/babelfish/trurl_pagecontent?lp=nl_en&url=http%3A%2F%2Ftweakers.net%2Freviews%2F621

[unhappy_mage]

What are you arguing? That RAID 0 makes no difference whatsoever?
That a single larger faster disk is more cost effective. Sure, the 740ADFD is faster than its predecessors. But dealing with cooling and powering a single 150ADFD disk is less difficult than two 74s.

The reason I argue against raid 0 is I see people with $1000 budget for a new machine going and buying a pair of Raptor 74s, when their budget would be much better spent on a faster video card or more memory. The 740ADFD is a great choice for this - about as fast as the 150ADFD - but it wasn't available at all until recently.

Sure, raid 0 gives more performance. But at at least twice the cost, and more than double the system complexity, not to mention the chance of catastrophic media failure. Thus, I recommend that people on budgets of even medium-large size stay away from raid 0. Does that explain it?

http://www.hardfolding.com/ftag1.php/mem/150072.png (http://www.hardfolding.com?go=38&tm=33&id=150072)http://www.hardfolding.com/utag1.php/mem/428/1.png (http://www.hardfolding.com?go=36&id=428&type=1)

[Brahmzy]

I see where you're coming from after you explained it a little better. However, pushing out gospel blanket statements as in previous threads is not advising people on what to buy within their budget, it was advising against RAID0 altogether. At least that's the way I read it. As shown in some of the benchmarks the new 74GB is as capable if not more in delivering blistering performance over the 150GB in single drive scenarios. However once RAIDed, the 150GBers pull ahead. At $500-$600, that seems to be extreme to me, but I am in no way going to advise against RAID for someone that has the cash to do it. It is simply faster. I have a complex backup scheme as it is, this will not change with RAID.

Anyways, onto something else. :)

[Adionik]

Do the rebates for these drives ever ACTUALLY come in?

I'd like to pick one up, but i'd really hate to get stiffed.

[EnderW]

And this is the OLD SCHOOL 74GB Raptors-not the new beasts. REAL WORLD testing here folks....

http://www.ocforums.com/showthread.php?t=445800
That "article" was shot full of holes a long time ago:

http://hardforum.com/showpost.php?p=1029199381&postcount=4
Okay, but seriously. He also claims:
- an increase of more than 100% for Photoshop loading times
- an increase of more than 100% for boot times
- 82% increase in speed with duplication of a "large" (1gb isn't large!) file versus only 45% increase with a 400 some MB file
- more than 100% increase in speed installing 3dmark

So how did the raptor benefit it more than 100%? Raid 0 fans, this thing has to be a disappointment - this guy's arguing in your direction but not getting believable results. This just doesn't add up.

http://www.hardfolding.com/ftag1.php/mem/150072.png (http://www.hardfolding.com?go=38&tm=33&id=150072)http://www.hardfolding.com/utag.php/mem/1392.png (http://www.hardfolding.com?go=36&id=1392)


And I could dig up about 4-5 articles that show little to no gain from RAID-0 on those same type of tests.
Here is one to get you started:
http://www.anandtech.com/storage/showdoc.aspx?i=2760&p=10

[mikeblas]

id like to see something other than a stopwatch being used.
What would you suggest using, and why?

[gman]

I don't even screw around with any benchmarks period. Never have.

But I do see a difference in RAID 0. My initial point was you can't tell me that 1 - Raptor 150 is faster than 2 - Raptor 74s RAIDed made from the same drive.

Now I do have a question about NCQ. On these Raptor drives will NCQ make any difference if it's not in some type of enterprise setup just Desktop?

[serbiaNem]

That "article" was shot full of holes a long time ago:

http://hardforum.com/showpost.php?p=1029199381&postcount=4



And I could dig up about 4-5 articles that show little to no gain from RAID-0 on those same type of tests.
Here is one to get you started:
http://www.anandtech.com/storage/showdoc.aspx?i=2760&p=10


The raptor tested by anand seems to be even slower than mine. On hdtach it gets 8.6ms access, mine gets 8.0, the avg seq is also 3mb slower. So those results might put the single 150 in an even better position. I was thinking about raiding two 150s, but I already have 5 drives in my case, and then I don't use any insanely disk intensive apps, so it wasn't worth it in my case. Plus, I probably have better access times than raid 0 arrays.

[Brahmzy]

Do the rebates for these drives ever ACTUALLY come in?

I'd like to pick one up, but i'd really hate to get stiffed.
They sure do. I got a $30 MIR from my first Raptor a year ago and it came within 3 weeks. Nice little check in the mail. Pick one up!

[Jason711]

What would you suggest using, and why?

if i could code, i would attempt to make something

[gman]

if i could code, i would attempt to make something

Why can't you code? I teach programming, it's not 'that' difficult.

[Jason711]

Why can't you code? I teach programming, it's not 'that' difficult.

a teacher saying it isnt that difficult.. ;)

[mikeblas]

if i could code, i would attempt to make somethingIf you could code, you'd know that it's pretty difficult to time boot up. By the time your code can run to start its timer, you've already booted the OS. How would you time the loading of a level in a game, for instance?

[Jason711]

your asking me how? you must be joking.

[mikeblas]

your asking me how? you must be joking.
Not at all. Since you wanted something other than a stopwatch, I figured you had studied the problem a little bit and had a reason for the request.

Am I giving you too much credit?

[Jason711]

if i knew how to code, perhaps i could understand it more. enlighten me, great one. :)

[drizzt81]

Not at all. Since you wanted something other than a stopwatch, I figured you had studied the problem a little bit and had a reason for the request.

Am I giving you too much credit?
I would say that the problem lies with that heisenberg uncertainty principle: Whenever you try to observe something, you change it. No different with the game loading times: without accessing and modifying their sourcecode, it would be virtually impossible to have a very accurate timing mechanism.

then again, I guess we could tap an image analyzer into the video output, and have an external hardware clock....

While I think that Jason's request was awkward, the bottom line is that the benchmarks shown by that guy on OCForums cannot be correct. Both you and I know that 1 + 1 is usually not more than 2.

[TeeJayHoward]

Both you and I know that 1 + 1 is usually not more than 2.
And yet 2+2=5 for extremely large values of 2! (http://www.thinkgeek.com/tshirts/science/60f5/)

Would it be possible to code something that compares the current process list with the number of I/Os in order to discover when a program is "finished loading"? AKA, start it when programx.exe is added to the list, and stop it when programx.exe has not submitted a single I/O request in 5 seconds. (Of course, subtract 5 seconds from the end time!)

[drizzt81]

And yet 2+2=5 for extremely large values of 2! (http://www.thinkgeek.com/tshirts/science/60f5/)

Would it be possible to code something that compares the current process list with the number of I/Os in order to discover when a program is "finished loading"? AKA, start it when programx.exe is added to the list, and stop it when programx.exe has not submitted a single I/O request in 5 seconds. (Of course, subtract 5 seconds from the end time!)
I guess filemon would be an option?

[Jason711]

would it be possible to put something in the bios to monitor from the time it initializes to o/s?

would it be possible for the game designers themselves to implement code inside of the game to monitor level loading? surely something can be running in the back ground to detect when, persay, a short cut is executed and when the program fully loads?

this is purely speculation, and i expect flames. *shrug*

[mikeblas]

I would say that the problem lies with that heisenberg uncertainty principle: Whenever you try to observe something, you change it.
I believe you're thinking of the observer effect, not the Heisenberg uncertainty principle.

I guess filemon would be an option?
Yes, but filemon is very resource-intensive. So we need something that is thinner, like Detours (http://research.microsoft.com/sn/detours/). I've used detours to study the file access patterns of a couple of games loading levels (http://www.hardforum.com/showthread.php?t=994470), and the problem is that the code is very poorly written. It ends up rereading data again and again; it's awful.

While Detours injects code, it's lots less invasive than FileMon. If the intercepting function ends up making a note of the transfer, it's just doing a little bit of memory access and not causing much overhead. FileMon ends up updating its UI and apparently isn't too efficient about storing the presented data, so it's lots more expensive.

Given a realistic method for monitoring the program, we still need a way to recongize that the load is starting (to start timing) and that the load is finished (to stop timing). Finding a start point from the log might be pretty easy --it's probably the first time a directory or file is accessed. Finding the end time is a little harder, since the game may or may not close the file when the level is loaded, might not read in the same pattern every time, and so on.

But I still don't understand why we should measure with anything other than a stopwatch. The results might be a little more accurate with a software-based solution, and just a little more repeatable. But not so much so that the results are different (given the numbers from that guy's thread). And we certainly avoid any problems inherent in changing the system we're trying to measure.

[drizzt81]

I believe you're thinking of the observer effect, not the Heisenberg uncertainty principle. yeah.. oops

Yes, but filemon is very resource-intensive. So we need something that is thinner, like Detours (http://research.microsoft.com/sn/detours/). I've used detours to study the file access patterns of a couple of games loading levels (http://www.hardforum.com/showthread.php?t=994470), and the problem is that the code is very poorly written. It ends up rereading data again and again; it's awful. I remember reading about this before. Maybe you made a post about this issue a while back?

But I still don't understand why we should measure with anything other than a stopwatch. The results might be a little more accurate with a software-based solution, and just a little more repeatable.

Was/ is it not you who wanted repreatable and reproducable results? (http://www.hardforum.com/showpost.php?p=1029554177&postcount=19)

-I am going to take my mouth to full now-

Wouldn't the option that you suggested be a good idea to use as a basis for a 'real-world' test? Not only would it be repeatable by the benchmarking person, but one could re-run the tests at home, provided that one has a copy of the game.

But not so much so that the results are different (given the numbers from that guy's thread). And we certainly avoid any problems inherent in changing the system we're trying to measure.
I still maintain that the numbers that the person in the OCForums thread has cannot be correct, since I fail to see how doubling the STR can derease load time by more than 50%. I am to close minded to understand that.

[HDBox2d1]

Diskeeper only takes 15 sec for a full defrag! Gotta like that.

Really? Hardcore.

[mikeblas]

I remember reading about this before. Maybe you made a post about this issue a while back?Yes, I did. Follow the link I posted; one of the ones that you quoted. I'm hoping to post a very interesting update to that thread this weekend, by the way.

Was/ is it not you who wanted repreatable and reproducable results? (http://www.hardforum.com/showpost.php?p=1029554177&postcount=19)

Using a stopwatch effects the repeatability of the measurement; not having enough information affects the ability to reproduce the results. They're two different kinds of "repeatability". (And maybe I've even used the wrong words, saying "repeatable" when I meant "reproducable".)

What I was asking for in the other thread was that StorageReview provide enough information so that we could have a crack at repeating the tests they did. The StorageReview tests aren't repeatable simply because we don't know, very specifically, what tests they did; the software they used isn't available, the capture files they use aren't available, and so on.

Here, we have the same problem: we don't know when this guy started and stopped the timer on a level load, for instance. We don't know which version of the drivers he was using, though we do have some information about his system configuration.

In both cases, we're hard-pressed to reproduce the guy's results. If we had the exact same hardware he did, we wouldn't be able to get the close to the same numbers without some guesswork about what he was timing.

If we had the same hardware that StorageReview uses, we wouldn't have any chance of getting the numbers they do because we don't have any of the software or data files they've got.

-I am going to take my mouth to full now-Sorry to be difficult, but I don't understand what that means.

I still maintain that the numbers that the person in the OCForums thread has cannot be correct, since I fail to see how doubling the STR can derease load time by more than 50%. I am to close minded to understand that.
You seem to be assuming that the single drive is running at its maximum throughput efficiency. What if it isn't?

[drizzt81]

Using a stopwatch effects the repeatability of the measurement; not having enough information affects the ability to reproduce the results. They're two different kinds of "repeatability". (And maybe I've even used the wrong words, saying "repeatable" when I meant "reproducable".) I think I understood you nonetheless.

What I was asking for in the other thread was that StorageReview provide enough information so that we could have a crack at repeating the tests they did. The StorageReview tests aren't repeatable simply because we don't know, very specifically, what tests they did; the software they used isn't available, the capture files they use aren't available, and so on.
[...]

I get your point and I applaud your strive for a scientific method of HDD testing that goes beyond HDTach. I am looking forward to seeing your updates soon.

Sorry to be difficult, but I don't understand what that means.
a poorly translated german proverb. don't worry. Something to the equivalent of "being in over one's head"... anyway.


You seem to be assuming that the single drive is running at its maximum throughput efficiency. What if it isn't?

*scratches head*
Let me see that I understand what you are saying: The single drive was using less than it's maximum possible STR of the section where the reads/ writes for that task were performed.
My knowledge about HDD's is limited, so I don't know if, when, why or how that would happen.Would it be related to the drive seeking more than transferring data?
Regardless of why the drive was running at less than "its maximum throughput efficiency", why should the RAID array be more efficient?

[mikeblas]

poorly translated german proverb. don't worry. Something to the equivalent of "being in over one's head"... I think translating idioms from language to language is funny. Will you tell me what the expression you meant was, in German?
My knowledge about HDD's is limited, so I don't know if, when, why or how that would happen.Would it be related to the drive seeking more than transferring data?
Regardless of why the drive was running at less than "its maximum throughput efficiency", why should the RAID array be more efficient?

Well, I know a thing or two about drives, but I can't guess why that would happen. I'm hesitant to offer a longer explanation because usually, when I do, people don't bother reading it and then assume I said what they wanted me to say, and then get on me about it.

So let's try this technique: I don't know anything about that guy's numbers. I don't know what happened, I don't know much about the tests he's running, and so on. Maybe he made 'em up. I ahven't investigated anything.

What I'm trying to address is the notion that we can throw is results out summarily just because they show more than a 50% improvement. I'm not saying his results are valid or useful; I couldn't say that until I actually reproduced them and measured them in detail and so on.

Is that clear?

If so, then let's think about why two things might run more than twice as fast as one thing. We're measuring how long some drive takes to transfer some data. We can model how fast it takes mathematically. The time it takes the drive to transfer a certian data size is that data size divided by the transfer rate of the drive:


time = (megabytes) / (megabytes_per_second)


If we want to compare two different timings, we divide them to get a ratio:


percent_improvement = (time2 - time1) / time2


Substituting, we figure that we really can't be more than 50% faster; but that only works if the drive is operating at the same rate. The difference between the rate we expect and the rate it works at is its efficiency. We really have:


time = (megabytes) / ((efficency) * (megabytes_per_second))


I'm too lazy to try to find a way to represent all the math in the forum's limited vocabulary, so let's go to a concrete example.

Say the raptor drives this guy is using are capable of STRs around 50 megs/second. When he tests a single drive, the drive runs at 40% efficiency. And let's say loading a level means transferring 10 megs. So,


time = 10 / (.4 * 50)
time = 10 / 20
time = 0.500 seconds


Say the RAID 0 setup has twice the STR, and the same efficiency:


time = 10 / (.4 * 100) // 100 is twice 50
time = 10 / 20
time = 0.250 seconds


indeed, it would seem we're limited to a max of 50% improvement. But what if the RAID0 setup is more efficient? What if it runs at 60% efficiency?


time = 10 / (.6 * 100)
time = 10 / 60
time = 0.167 seconds


We're suddenly more than 50% faster than the drive.

Again, I'm not saying this is what happened for this guy's test. But it certainly is possible that two drives run more than 50% faster than a single drive, and so it's silly to say that the reuslts should be immediately discounted.

Of cousre, the natural question is: what could cause that? And did that really happen here? Well, I can't say. It doesn't seem likely, but I haven't studied his setup and tests (and can't, because there's information missing).

On the other hand, there's very obvious and simple cases that are common and normal when this kind of thing happens. The ones I know about don't involve RAID 0. Say you have a database program that wants to write both to the database and also to a transaction log. You've put both the database and the transaction log on the same disk, and start inserting lots of rows of data to your DB.

Well, the head is going to be thrashing around: you write a little to the database file, then seek to the transaction log file, then back to the database file, and so on. This makes the effective efficiency pretty low.

But then you go to the store and buy another drive, and move the transaction log file to that drive. You have good hardware and know how to handle your cables, and so the database software can issue concurrent requests to both drives at the same time. The I/Os actually end up being serial in both cases; seeking is minimized, and so on. This is far more efficient, and there's no contention anymore, so your two-disk setup ends up with better than 50% improvement compared to a single drive.

Can something similar happen between a single drive and a two-drive RAID0 setup? I don't see a path where it can, but one might exist. You ask: "Why is the raid0 array more efficient?" and I don't know. But if it is, it's entirely possible that he gets better than a 50% improvement.

[drizzt81]

nice post. I understand what you are saying.
Thinking about it some more and throwing out my anti-R0 biad for a minute:

Say we wanted to load a bunch of data (A,B,C) that is located on different areas of the disk, i.e. bad locality. While A and C are 'near' each other, C is far away. Each chunk of the data is smaller than a stripe (i.e. is completely located on a single disk in the two-disk array). On a single disk, the read head would have to move to chunk A, read it, move to chunk B (which is 'far away' from A) read that and then move to chunk C and read that, which means that there is a lot of head movement involved.

On the RAID array A and C are located on disk 1, close to each other and B is located on the other disk. If the RW-heads were positioned so that head 1 could read chunk A in parallel with head 2 reading chunk B, then there would be an efficiency improvement.

While this sounds plausible, I think it is rather unlikely, since such a head-placement would only occur if the two disks are 'out of sync'. What I mean by that is that if we wanted to tranfer chunks A and A+ (which are in sequential stripes) we'd have to wait for the other RW-head to move into the correct position before being able to complete that transfer. However, I do not know how software RAID-0 implements head-alingment or whether the heads are intenionall placed at 'opposite ends' in order to reduce the time until 'something' is being read.

Last but not least, I don't know how likely it is that the data happens to be fragmented in such a way that this effect occurs. The bottom line is: anything that you or I speculate about the tests shown my the person on the OCForums is just that speculation.
I guess it'd be nice if we had that reproducable result (*phew* talk about coming full-circle) which we could analyze and verify.

[mikeblas]

Last but not least, I don't know how likely it is that the data happens to be fragmented in such a way that this effect occurs. The bottom line is: anything that you or I speculate about the tests shown my the person on the OCForums is just that speculation.
I guess it'd be nice if we had that reproducable result (*phew* talk about coming full-circle) which we could analyze and verify.
Indeed, we don't know, and we're stuck guessing. If people think this guy's post has been "poked full of holes", they should go over there to that other forum and ask him about those holes to his face, in the thread, and give him the benefit of a peer review. That's helpful and productive; either he learns something from his mistake, or we learn something about how to better setup RAID 0 systems.

You now know why I want to see reproducable tests. If I don't agree with the results (or even if I do!) I can try them out and measure it myself. I can convince myself of the results, and in doing the exercise, I probably learn a lot of interesting things.

Anyway, I was also thinking of your same analysis. Maybe the data is interleaved exactly ideally, and the the RAID array (the controller, I guess) can overlap the requests when it knows they're drive-independent. This would really help the problem that people think RAID0 has with latency and "small file I/O", and would be an interesting optimization to pursue.

(Then again, I don't even know if two concurrent I/Os are possible on montherboard-based RAIDs, even to different devices. Wouldn't they both want to exclusively access the bus for the DMA operation when the drive started returning the requested sectors?)

While that physical disk layout is possible, I don't think it's probable--it's not very likely to happen on a new install, which is what this guy said he used. Plus, for it to add up to a meaningful difference, it would have to be true for lots of I/O operations in a row, for a long time. That makes it seem even less likely.

<shrug> But as you point out, we're just speculating.

[drizzt81]

Indeed, we don't know, and we're stuck guessing. If people think this guy's post has been "poked full of holes", they should go over there to that other forum and ask him about those holes to his face, in the thread, and give him the benefit of a peer review. That's helpful and productive; either he learns something from his mistake, or we learn something about how to better setup RAID 0 systems.

excellent point:
http://www.ocforums.com/showpost.php?p=4550511&postcount=75

[mikeblas]

excellent point:
http://www.ocforums.com/showpost.php?p=4550511&postcount=75
good start! I'll try to keep an eye over there to see what the responses are like; it seems like nobody in that forum has asked any critical questions and they're all syncophantic. But please let us know if you get any good explanations.

[EnderW]

good start! I'll try to keep an eye over there to see what the responses are like; it seems like nobody in that forum has asked any critical questions and they're all syncophantic. But please let us know if you get any good explanations.
lol, that forum is a joke

[Jaap74]

just tried to read this forum............

Zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzz

it was good for the first few pages............

the people tried to score points off eachother..............

:o
:(
;)

[Dan_D]

I am going to grab 15K U320 Fujitsua MAU drives or something next personally, but yes those Raptors should kick ass.

[drizzt81]

just tried to read this forum............

Zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz zzzzzzzzzzzzz

it was good for the first few pages............

the people tried to score points off eachother..............

:o
:(
;)
This Forum ([H]Forums), that Forum OCForums or this thread?

On a side note, I even PM'd the Dan person, who did the benchmarking asking him to reply, but it appears that I have been ignored so far.

[gman]

This Forum ([H]Forums), that Forum OCForums or this thread?

I think he means the thread. I think you guys are possibly talking over some heads here or the discussion has wavered from the main topic or simplied topic.

I personally enjoyed your discussion and learned some things. Thanks.

[mikeblas]

I personally enjoyed your discussion and learned some things. Thanks.Me, too!

[drizzt81]

quick update: I just sent my second PM to Dominick32 and I have yet to receive a reply or a pm from him. Very frustrating. I should have the hardware to replicate his experiment in the middle of next week. Now, if I would only know what to measure...

[general]

"What are you arguing? That RAID 0 makes no difference whatsoever?

I know the gains may be small in some areas, but I don't care. Don't dump on RAID because you (with your particular budget) don't think it's "worth it." If my PC boots up onto the desktop 3 seconds faster, it's worth it to me. If BF2 takes 5 seconds less to load a map, it's worth it to me. Nuff said."

I guess I just hope it is worth the dropped framerate due to the CPU utilization of the RAID. As before, the optimal solution is just to get the bigger raptor. If budget is no option, then go with striping and mirroring and the discussion is null. Then again, if budget is no issue, you'd probably be looking at 15k SCSI drives.

[mikeblas]

We'll just have to go with our guesses. What hardware will you use?

[drizzt81]

We'll just have to go with our guesses. What hardware will you use?
as soon as I get my stuff in, I will have a 3000+ Venice on an A8N-E with a pair of older Samsung 160GB drives. Even if my results are crappy in terms of performance, I think that my goal is to create a reproducable benchmark/ test by being verbose about my methodology. Unfortunately, the only paired drives I have are these Samsungs and I am not willing to fork out the big bucks for a set of raptors.

[drizzt81]

we have a resposne:

http://www.ocforums.com/showthread.php?p=4561039#post4561039

Anyway, the Windows XP boot was timed immediately after the BIOS Posts its final figures (ex: Hard Drive Initialization or Raid Init) and is completed when the XP Professional login prompt appears and the hourglass dissapears.

[drizzt81]

*bump* since this topic has come up again.

[mikeblas]

Huh? Where did it come up?

[TeeJayHoward]

Huh? Where did it come up?
Here (http://www.hardforum.com/showpost.php?p=1030118117&postcount=42) in this (http://www.hardforum.com/showthread.php?t=1111888&page=3) thread.

[codegrinder]

In any case raid is better depending on what you use it for as stated above in numorus posts. I am able to burn 2 dvds @ 16x at the same time and that is something I cannot do with them not raided together.

[unhappy_mage]

In any case raid is better depending on what you use it for as stated above in numorus posts. I am able to burn 2 dvds @ 16x at the same time and that is something I cannot do with them not raided together.
16x dvd is only 24 MB/s, it's not a huge accomplishment. And in any case, you'd be better off with two seperate drives, one each with one ISO image on it. I have done tests to this effect.

http://www.hardfolding.com/ftag1.php/mem/150072.png (http://www.hardfolding.com?go=38&tm=33&id=150072)http://www.hardfolding.com/utag1.php/mem/428/1.png (http://www.hardfolding.com?go=36&id=428&type=1)