![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
RAID-0 proved ineffective at boosting desktop application/game performance
- Thread starter DougLite
- Start date
unhappy_mage
[H]ard|DCer of the Month - October 2005
- Joined
- Jun 29, 2004
- Messages
- 11,455
That's what my tests did, essentially. Reading from /dev/zero is a special device that returns ascii NUL (0x00) bytes for as long as you read it. What the test did is copy 25 GB of /dev/zero into a file on disk, and then duplicate that file. As the tests showed, even in creating the file RAID 0 didn't help.Lord of Shadows said:
I'm working on a driver issue with the pci ide card, but once that's over I'll post results.
You can copy faster in RAID 0, and theres a billion charts to prove it on the net. Whats wrong with that?
Everyone knows farcry load times suck in RAID, but I'd chalk it up to programming that didn't take into account RAID.
So you miss out on 5 seconds of farcry load, but you save hours of copying files around your computer.
Everyone knows farcry load times suck in RAID, but I'd chalk it up to programming that didn't take into account RAID.
So you miss out on 5 seconds of farcry load, but you save hours of copying files around your computer.
rogue_jedi
Supreme [H]ardness
- Joined
- Sep 6, 2002
- Messages
- 4,821
Time for disk to disk copy of 25gb: 10:59 (37.9mb/s)provoko said:
Time for raid0 to self copy of 25gb: 30:29 (13.7mb/s)
setup used:
dual 933 p3 with 1gb memory, xp sp2
SIIG PCI ultra ata 133 card
2x ST3120026A on one channel (tested hdtach on one disk while extracting to the other, performance loss over hdtach on one disk w/o the winzip was negligible - 2mb/s burst speed loss)
windows xp dynamic disks for raid 0, seperate dynamic partitions for disk to disk
timing was performed with a stopwatch, standard drag-and-drop copy. if you want a "better" test, let us know.
a 25gb file was generated in linux, gzipped, and transfered to the xp box. unzipped to either one of the disks or the raid0, depending on the test. for the raid0 test, i simply said "copy here" for the file. i think the results speak for themselves; copying a file from one disk to another physical disk is 3(!) times as fast as going from a raid0 to itself.
tests performed by myself and unhappy_mage.
we are planning to rerun the tests under linux in order to get a more accurate time, but we thought people might like to see the XP results before i reformat and reinstall.
rogue_jedi said:
So that proves that software raid0 (not even cpu assisted hardware through it's driver) is slower. Some of us knew that already, and does nothing to prove much of any point other than people will use a poor benchmark to prove a point.
unhappy_mage
[H]ard|DCer of the Month - October 2005
- Joined
- Jun 29, 2004
- Messages
- 11,455
You want to lend me the other two disks needed? The costs of raid 0 are higher, remember?Ockie said:
The second result is copying from a raid 0 array to itself. Is that what you meant?
DougLite said:
Holy shit, go away for awhile and look what happens. DougLite, you can change the title, it still makes no sense. Im going to say this once, and then im going to bow out of this one. You have a hatred for RAID0 period. End of story, no denying that. And ill tell you why. First RAID0 is dead. Then the topic becomes gaming oriented. Next desktop usage and gaming. Well I will gladly admit, as I have before that I agree with you from a gaming standpoint. But as far as desktop usage goes, and this is the main point, your own statistics YOU provided in the first post shows the opposite of what you claim to be fact. Plain and simple you "debunked" your own belief ! Your personal crusade against RAID0 was lost when you posted Eugenes benchmarks. Office DriveMark and High-end DriveMark, in which both consist of MANY desktop apps., show RAID0 doing is job quite nicely. And thats a fact you simply cannot deny! RAID0 shows a benefit in desktop apps.
This isnt the only time you seem to ignore your own studies/tests. There have been a couple forum members who have pointed this out in the past on this subject. Just offhand I think one was MikeBlass(spel.?). I saw the comment you made regarding some people may hate you for your belief. I dont hate you DL. But I do SERIOUSLY question your motives. I do not know if your trying to boost your "social standing" in the forum, or what it may be. Truthfully there have been a couple times I have questioned in my mind your authority as a Moderator. You come off too extreme sometimes, whether it has to do with this topic or whenever you lock a seemingly harmless thread. This is my opinion, take it as you may, it is not meant to be an arguement.
Masher, as far as your concerned, keep digging yourself a hole. Thats the most entertaining part in this thread.
For others who may not know what to believe, respect DougLites wisdom, he is very VERY knowledgable. But please, before you make a decision on this subject, read the studies, look at the numbers, dont just listen to what we say. Do your homework, then decide. Take care all.
EDIT**
PS. I just want to thank unhappy_mage for personally taking the time to do some tests. I appreciate that. Take care.
rogue_jedi said:
Okayyyyyy, here you go:
http://www.techreport.com/reviews/2005q4/chipset-raid/index.x?pg=7
And:
http://www.techreport.com/reviews/2005q4/chipset-raid/index.x?pg=9
The charts clearly show copying and creating in raid0 is faster vs a single drive.
I don't know what you did wrong, but theres a billion other websites that already have proven raid0 can copy faster than a single drive.
unhappy_mage
[H]ard|DCer of the Month - October 2005
- Joined
- Jun 29, 2004
- Messages
- 11,455
We're doing a different test than that. We created a file 25gb in size (not timed) and then either:provoko said:
a) copied it from drive a to drive b or
b) copied it from raid 0 array composed of drive a and drive b to the same volume.
So it's apples to oranges. The original question (I thought) was whether situation b was faster than a; PaHick has since changed his statement to mean copy from a raid 0 to a single drive. I agree, in this case it could be faster. But you can see from my (honest, real-life, replicable) results that raid 0 is *not* a magic cureall.
That's my point; raid 0 doesn't magically make things fast. You have to plan out your system (especially on a budget) in order to get decent speeds, and one of the things you have to take into account is what you'll be doing with the system. Simply pointing people to raid when they complain about speeds is absolutely the wrong thing to do, and that appears to be the mindset in this thread.
I just don't know what to say to this. Software raid is no slower than hardware; levels with parity (3, 4, 5, etc) are implemented with XOR chips and such, 0 and 1 don't need that. They're just taking the same blocks and either alternating which disk they get sent to or writing them to two disks. This is probably limited by the speed of copying blocks of memory. There's just no reason for speed differences between "hardware" and "software" raid 0. If you can show me a benchmark where the same card scores markedly differently in hardware versus software, I'll do my best to eat my words, but I'll be surprised if you find one.matguy said:
PaHick: Do you have any other suggestions for good benchmarks to run? I'll be glad to run them and post the results, regardless of what they show.
Good god, can people not read? He's not testing a single drive, he's testing a 2-disk Raid 0 array against 2 independent drives. We only said this about 30 times throughout the thread.provoko said:
You are correct. Well, to be precise, the last time I saw a benchmark, software Raid 0 was very slightly slower than a dedicated card, but it wouldn't even knock the hat off the 300% speed differential Jedi's test showed.unhappy_mage said:
As I already said, it doesn't even take a benchmark to realize two independent disk should be much faster than a Raid 0 for file copies...it just takes a tiny amount of common sense, coupled with an understanding of how discs work.
Benefit? Ha! WD1500ADFD matches or defeats 2x or 4x WD740GD in RAID-0 in all five of SR's single user tests. True, running WD1500ADFDs in RAID-0 would deliver some modest scaling over a single WD1500ADFD in 3 of the 5 DriveMarks, but double cost for a gain of 10% or less (that isn't even across the board) doesn't make sense to me. Show me an "upgrade," any upgrade, that is less cost effective.PaHick said:
I'm not ignoring them, I'm dismissing them
because the gains aren't cost effective, once again in my judgment
Kyle already thinks I'm nuts
and I'll never be a super mod or admin. So what?The tests you link to are incomplete/ inaccurate. They do not show a test copying from Drive A to Drive B. They compare single drive to RAID-0, which is an incorrect comparison, since RAID-0 requires 2/3/4 times the money investment up front.provoko said:
so there must be a Billion-1 examples that show what you claim. Maybe show us another one?
ps.: thanks UM for running the test.
drizzt81 said:
WHY WHY WHY are you people testing such a stupid scenario AND ONLY ONE just ONE scenario for that matter? Soooo because raid sucks in one aspect it sucks in all aspects? What is everyone's problem here?
What a bizarre discussion, it's not like !!!!!! talk like ati vs nvidia or amd vs intel. It's everyone's misconceptions vs misconceptions.
All I'm saying, and has been proven, yes a billion times (use google), that raid0 is faster in reads and writes. No, not in all situations, SHOOT ME, it can't copy to my external drive as fast as a zip floppy can copy to the cdrom dvd rewritable, or it's i/o's are faster on munix than lindows and it uses 1 watt more than a single vibrator, no, not in all situations indeed, so I'm sorry.
This is a biased discussion and should be closed. Any further discussion should be in it's own thread with it's own point so that you can discuss your particular picky aspect of raid and knock it or praise it.
Whyyyyy are you people even wasting your time, crap like this has been discussed already.... I'm ending my subscription to this thread, hahaha.
Agree to the above!!
Your tests are inaccurate and your excuse is weak.
unhappy_mage said:You want to lend me the other two disks needed? The costs of raid 0 are higher, remember?
The second result is copying from a raid 0 array to itself. Is that what you meant?
Your tests are inaccurate and your excuse is weak.
Instead of screaming wildly, why not read the thread and find out? We tested this particular scenario because a Raid phanboi claimed it would outperform two disks in this scenario. Proven false, btw.provoko said:
No, we've discussed, analyzed, and tested several other scenarios. Read the thread.
Say all you want. What's been *proven*, however, is that for desktop users, Raid 0 is one of the cases below:
1. Slower than a non-Raid configuration
2. Significantly faster than non-Raid...but in a scenario that gives zero real-world benefit
3. Very slighter faster than non-Raid in a scenario that benefits a user...but by such a tiny fraction that the additional failure rate, cost, and complexity is not worthwhile.
4. Faster than non-Raid in a beneficial, worthwhile manner...but in an extremely rare and atypical scenario most users will never experience.
Lord of Shadows
2[H]4U
- Joined
- Mar 9, 2004
- Messages
- 3,322
I really dont see much proof on the first post. The 2xRaid compared to the 4xRaid shows and improvement, which is in direct contrast with your "logical" conclusion. This thread isnt 7 pages deep because of hard undeniable results.
To say raid doesnt improve performance at all is indefensible.
To say raid doesnt improve performance at all is indefensible.
The ability of people to misunderstand simple English is truly amazing. No one said anything at all like this. Read my post above to say what *is* being said.Lord of Shadows said:
As for this thread being 7 pages long, its primarily for the reason demonstrated by your post...people with shoot-from-the-cuff opinions that don't have any relevance to the discussion. Raid is obviously faster in some cases. It's just not faster in ways the average user garners benefits from.
Lord of Shadows
2[H]4U
- Joined
- Mar 9, 2004
- Messages
- 3,322
DougLite said:
Not to mention all the other flawed test results still on the first post, you know the one people actually read.
You keep speaking of proved results, but Im just not seeing it.
I'm speaking of the StorageReview benchmarks linked to...as well as Jedi's own informal test...not that you even need a test to prove that last bit. So far we have the following scenarios we've looked at:Lord of Shadows said:
-> Gaming results: slower on many games, slightly faster in others. Net wash, certainly not worth the added complexity. Put your money into more ram or better cpu/gpu.
-> Office apps: faster...but who wants to trade a doubled chance of losing a spreadsheet or document in exchange for Word loading a few percent faster? These apps are already far faster than they need to be, and idle 99.9% of the time.
-> Video capture: device bound, no performance gain.
-> Video transcoding: cpu bound, no performance gain.
-> File-file copy: far, far faster on 2 disks than Raid 0.
If you want to analyze or posit benchmarks for any other scenario, please do. I'm sure I'm like most people here...Raid in a desktop is just so damn _cool_ that we'd love to be proven wrong, and it to have some real benefit.
Lord of Shadows
2[H]4U
- Joined
- Mar 9, 2004
- Messages
- 3,322
Your farcry results are flawed because farcry has been proven by mikeblas to not make efficient use of the drive when loading. Not to mention the whole deal about varying file sizes with stripe size I mentioned a while back etc.
Office apps are irrelevant completely by your logic, why buy a faster single drive then?
As for the reliability, I think your playing that a bit hard. Yes you lose everything if you lose one drive in the array. If my drive lasts four years, raiding it with another equivalent drive doesnt suddenly make either die any sooner. Having a non-raid setup doesnt save your files if your main drive goes down. Duplicating files is still important whether you're raided or not.
I can agree with a bottleneck of a device, but again a faster single drive doesnt help there either.
As for file transfer, your test is flawed. You copied a file from one drive to another, while raid copied from one drive to itself, which introduced countless seeks from position A to B as buffers filled.
Now your working at a higher level than you need to with your testing, you can do two interesting things with raid. You can read/write large blocks of data, which can be proven with any drive benchmark out there and has been. And raid should also help a little (I assume seeking would be the largest factor here) when transferring a lot of small files with a small tripe size. I would like to see these numbers for myself though.
So raid is about as worthless as continuous read/write speeds.
Office apps are irrelevant completely by your logic, why buy a faster single drive then?
As for the reliability, I think your playing that a bit hard. Yes you lose everything if you lose one drive in the array. If my drive lasts four years, raiding it with another equivalent drive doesnt suddenly make either die any sooner. Having a non-raid setup doesnt save your files if your main drive goes down. Duplicating files is still important whether you're raided or not.
I can agree with a bottleneck of a device, but again a faster single drive doesnt help there either.
As for file transfer, your test is flawed. You copied a file from one drive to another, while raid copied from one drive to itself, which introduced countless seeks from position A to B as buffers filled.
Now your working at a higher level than you need to with your testing, you can do two interesting things with raid. You can read/write large blocks of data, which can be proven with any drive benchmark out there and has been. And raid should also help a little (I assume seeking would be the largest factor here) when transferring a lot of small files with a small tripe size. I would like to see these numbers for myself though.
So raid is about as worthless as continuous read/write speeds.
rogue_jedi
Supreme [H]ardness
- Joined
- Sep 6, 2002
- Messages
- 4,821
duplicating files is always important, the real raid configurations do this for you (to some extent).Lord of Shadows said:
while it is true that adding another drive and putting them in raid 0 won't kill your original drive any faster (unless you have a really bad PSU or something like that), you now have twice as many drives in which a failure can occur. it is more likely to have a drive fail on you if you have 2 than if you have 1.
i realize this; as stated i don't have another pair of disks available to test with. i still don't get why this is not a "valid excuse" - we tested what we set out to test. PaHick has since amended what he said, adding another drive to the equation. in that case, he still gains nothing from raid 0.Lord of Shadows said:
why? because the raid 0 can be as fast as he can make it. you are still only going to get what the single drive can push. it makes sense if you have 1 new disk and 2 old disks, but with all new or all old disks it makes no difference. the only other drives in the test machine are on ATA33 and are an 8gb and a 10gb drive, both 5400 and old/slow/fragmented/slow/etc. so doing that test would likely "prove" raid 0 to be equal in speed to a single disk, +- a negligible amount of course.
i decided to time another test - this time, in response to your criticism about being from one disk to another as compared to the raid 0 to itself, i am copying the file from a single disk to itself. i am running the test as i type this, and i have no idea how it will turn out.
so far it appears to be about the same speed as the raid0 to itself, but i do not trust windows progress bars. so we'll see in about 20 minutes.
edit: results!
Time for disk to disk copy of 25gb: 10:59 (37.9mb/s)
Time for raid0 to self copy of 25gb: 30:29 (13.7mb/s)
Time for disk to self copy of 25gb: 30:37 (13.6mb/s)
i think that pretty much speaks for itself... (which is good, cause i don't feel like writing anything about it)
the black knight always triumphs!unhappy_mage
[H]ard|DCer of the Month - October 2005
- Joined
- Jun 29, 2004
- Messages
- 11,455
1) Okay, let's revive that discussion. Test all your games and report back, and we'll be glad to benchmark raid-0 versus single disk setups.Lord of Shadows said:
2) OMFG how many times do we have to say this?
masher said:
3) Okay, sure. What's a small file, and what's a small stripe size? How's this distribution for a small file set:
27878 files in 1443 directories, 512289K disk space used, actual size of files 455107K. The average file size is ~16k.
rogue_jedi said:
For the record, I didnt amend anything. I meant that all along. And I did gain speed. File transfers from the RAID0 to another drive is faster. I dont know what you guys are doing, it could be a hardware situation causing this I dont know. I know what MY machine does. I know what tests have proven, and that is file transfers benefit from RAID0.
DougLite said:
You totally dismiss what your saying is fact. Your title...RAID-0 proved ineffective at boosting desktop application/game performance. Nowhere does it say, or do you imply it says "RAID0 cannot beat newer generation drive!". Your implying that RAID0 sucks altogether. The truth is this, and what I have been saying all along, is that compared to a single-same generation drive, RAID0 is faster. Really quite simple isnt it? But instead of sticking to what you type, as in RAID0 shows no benefit to the desktop user, you run on and on in so many different directions....its not faster against a newer generation drive, its not this, its not that. The fact remains that Eugenes benchmarks(BTW, not yours) prove against a simliar drive, RAID0 wins. Just because YOU dismiss them does not make that a fact! You can continue to post BS and tests and then contradict yourself. And it is time to bow out for good this time..lol. Please continue with your discussion.
I agree with you that in most cases, RAID0 will be faster. Significantly faster? Well that depends on your definition of signicant and what benchmark/test you are running.PaHick said:
Gaming sees virtually no benefit, file transfers do see a big improvement, but usually similiar (or greater) gains could be accomplished with seperate disks. Overall desktop performance does seem to increase some as seen in the benchmarks in the first post and this test from 2004:
http://faq.storagereview.com/tiki-index.php?page=SingleDriveVsRaid0
I see it similiar to SLI, just with much smaller gains. Unless you have everything else maxed out on your system, there is almost no reason for it. A Raptor 150GB RAID0 is about the only ATA RAID0 setup I'd consider since a single Raptor 150GB is fast enough to outpace the modest gains of setting up RAID0 with any other ATA drive.
Kinda like how a single 7800GTX is faster than two 6800GS in SLI, but two 7800GTX in SLI, while impractical for most situations, is still faster than a single one.
rogue_jedi
Supreme [H]ardness
- Joined
- Sep 6, 2002
- Messages
- 4,821
well, that isn't what you said at first...PaHick said:
you mentioned nothing about using a single drive; you just said "duplicating". we took that to mean from the raid 0 to itself, which seems reasonable to me, based on that statement.PaHick said:
wrong your/you're, too
just curious, but when you do your video editing, do you transfer it to the raid0 and then just work on it and save it there? i'd think you would lose any benefits from raid0 because you are now copying from the raid0 to itself, which is where seperate disks shine...
wrt the "from the raid0 to another drive is faster" i'd like to know your basis for that statement. i'm guessing you're still limited by the single disk, and raid0 is gaining you nothing.
Eugene was merely refuting the idea that "the 150gb raptor sucks cause i can get 2x 74gb raptors for the same price and raid 0 them, and it'll be faster" when he did his benchmarks, at least according to the thread. funny how many people there, even knowing what he said exactly, basically said "stfu noob" for saying a single 150gb raptor is better than 2x 74gb raptors in raid0.
Your logic is flawed. It doesn't matter WHY any particular game doesn't benefit from Raid-0. What matters is the benefit doesn't exist.Lord of Shadows said:
Some games do have a very slight benefit from Raid. Some nonel. Some run slower. Net result: a wash. For the scenario of "I'm-a-gamer-what-do-I-do-to-get-more-performance", the end result is the only thing that counts.
You don't, of course. Which if why users primarily of Office apps don't buy Raptors either. They're usually even happy with 4200 rpm laptop drives. Raid 0 for these users is a joke.
Flawed logic. Let me give you an analogy to explain why. Let's say you built an Raid 0 array that didn't have just two drives, but two thousand. Each of those drives has a MTBF of 40 years. That doesn't mean each drive lasts 40 years...it means you get 1 failure per drive per 40 years. That means 2000 drives experience an crash once per WEEK. All the drives are highly dependable...but your array crashes once per week.
Thats statistics, it doesn't mean it happens like clockwork. One week you might be fine...then the next week you have two crash at once. But over the long term, you'll AVERAGE one drive failure a week.
Going back to our two drive array. It has twice the chance to fail as a single disk. Actually, slightly more than half, as interdisk sync errors are possible. Even comparing that two disk Raid to two separate disks, it loses on reliability....because although both setups will crash about as often, the independent disks only lose HALF your data when they do. The Raid array loses everything.
Of course. But no matter how good your backups are, you have to assume a lengthy loss of time for every crash, plus loss of everything you did since your last backup. So why increase the odds without good reason?
Glad to see you're getting the point here. If a faster disk doesn't help, why bother with Raid? Stick with a single disk and put your money elsewhere.
Oops, the test was not flawed. For our scenario of "I need to duplicate video files on a regular basis" thats just how you'd do it. You'd add a second disk and copy from A->B. Thats the whole POINT of adding a second disk. So it'd be faster than just copying the file from one point on the disk to another.
See the point? If you're doing large file copies, you can speed it up by either a Raid array...or a second independent disk. But the Raid array is slower, more complex, and less reliable than the indie disk approach. So why choose Raid?
unhappy_mage
[H]ard|DCer of the Month - October 2005
- Joined
- Jun 29, 2004
- Messages
- 11,455
Test description:
Copy a kernel .git tree (mostly small files, two large ~100mb files) from a ramdisk to either a raid 0 or a single disk.
Results:
Raid 0: 173 seconds
Single disk: 187 seconds (8% slower)
So 200% the cost, and 8% faster. Any other tests?
Copy a kernel .git tree (mostly small files, two large ~100mb files) from a ramdisk to either a raid 0 or a single disk.
Results:
Raid 0: 173 seconds
Single disk: 187 seconds (8% slower)
So 200% the cost, and 8% faster. Any other tests?
Lord of Shadows
2[H]4U
- Joined
- Mar 9, 2004
- Messages
- 3,322
@masher: Those games wouldnt benefit from a faster single drive either, thats like saying I shouldnt raid because it doesnt speed up my downloads.
Im not talking about average reliability, I can guarantee you that if you had 2000 drives, they would all last longer than a week, they might all die in the same distant year, but they wont die every other week like your making it sound.
And I dont know many people that copy large files on the same drive for any reason, but Im sure anyone would appreciate the extra boost when loading say a video file into memory for editing.
We can all agree that raid doesnt double the speed of all drive operations, so instead of just saying raid is good/evil perhaps we need a sticky that explains how it works and not how we feel about it.
Im not talking about average reliability, I can guarantee you that if you had 2000 drives, they would all last longer than a week, they might all die in the same distant year, but they wont die every other week like your making it sound.
And I dont know many people that copy large files on the same drive for any reason, but Im sure anyone would appreciate the extra boost when loading say a video file into memory for editing.
We can all agree that raid doesnt double the speed of all drive operations, so instead of just saying raid is good/evil perhaps we need a sticky that explains how it works and not how we feel about it.
None of you guys have met my challenge, except for one guy who posted the joke about case lighting: Show me a computer upgrade, any upgrade, that is less cost effective than "upgrading" to RAID-0. Show us all. Make me wrong.
Even SLI and Crossfire show better return on investment than RAID-0. Even a Pentium Extreme Edition is a better upgrade...at least it improves performance in everything.
Even SLI and Crossfire show better return on investment than RAID-0. Even a Pentium Extreme Edition is a better upgrade...at least it improves performance in everything.
Lord of Shadows
2[H]4U
- Joined
- Mar 9, 2004
- Messages
- 3,322
I bought two sata drives on blackfriday for $20 bucks a pop and have them in raid0, I dont see where your argument about cost efficiency is coming in. Later, if I needed a drive for some reason I could easily take one out, an extra drive doubles my storage capacity and lets me double my peak bandwidth for "free".
True, but one thing to consider with CPUs and video cards, often you can overclock them to reach the same level as the more expensive chips. You can't overclock a hard drive, so once you've bought the single fastest drive that's out, the only thing left is to buy another.DougLite said:
As for your challenge, I'll try
Say I had a Raptor 1500 with my current setup of a Opteron 170 @ 2.6GHz. I could buy the new FX60 for about $1300, and sell the 170 for about $400, so the upgrade would cost me $900. Now, the FX60 runs at 2.6GHz stock, but it's unlikely I'd get more than 2.9GHz out of it with air cooling.
Let's also assume that performance scales linearly with clock speed, so I should see an overall increase of about 11.5% at a cost of $900.
Or if you want to take overclocking out of the equation, let's compare the $1300 FX60 at 2.6GHz to a $500 X2 4400+ at 2.2GHz. Then it's a 18.1% increase for $800.
Or I could buy another Raptor 1500 for $300, that should give me about an 8% increase in SR's 2002 Office and High-End Drive Marks seen here.
Of course, you see gains in different (and fewer) areas with RAID0 than you would with a CPU upgrade.
Some other things to consider:
-------------------------
Part of the performance increase (if it does occur) in RAID-0 stems from simply the increased capacity of the logical volume, intensifying the effects of locality and reducing seek distances...an effect you can achieve by simply buying a drive that is twice as big. This is why running a single 250GB drive is better than running a pair of 80GB drives in RAID-0...you get the benefit without any of the drawbacks of RAID-0, for the same cost, while achieving more capacity.
-------------------------
Also, running RAID-0 increases the impact of rotational latency, as the spindles are not in sync. Rotational latency is in average. For a 10K drive, that average is 3ms. It takes 6ms for a platter to make a complete rotation at 10,000RPM. Sometimes the actuator will get in position for a request, then the data on the platter will pass underneath the heads fairly soon, say around 1ms. Other times, the heads may "just miss" the requested sector, and the drive may have to wait as 5ms or more for the platter to spin back around.
Let's look at the impact of rotational latency under a purely random load. If you are running SLED, 50% of the requests have a latency above 3ms, 50% below 3ms. Therefore it averages out. However, when you are running RAID-0, you have two spindles that are not in sync. Presuming that the requests "take advantage" of both drives, the entire point of running RAID-0, let's look what happens:
Drive 1 may have a < 3ms latency, and drive 2 may have a < 3ms latency. In that case, the rotational latency on that request is below average. If drive 1 takes > 3ms, but drive 2 takes < 3ms, the system may have to wait for both drives to complete their request, as the requested data may need to be processed to determine what to load next. In that case, the rotational latency is above average. Drive 1 may be < 3ms, but if drive 2 is > 3ms, you are once again faced with above average latency, and (obviously) if both drive 1 and drive 2 are > 3ms, then your rotational latency is above average.
Let's chart this up to make it clearer
SLED:
< 3ms: low
> 3ms: high
RAID-0:
< 3ms < 3ms low
> 3ms < 3ms high
< 3ms > 3ms high
> 3ms > 3ms high
When running SLED, 50% of the requests have an above average rotational latency, but when running RAID-0, 75% of the requests that require both drives to be serviced (isn't the goal of running RAID-0 to have two or more drives working on requests?) have above average rotational latency, thusly the performance decrease in the seek performance limited tests FarCry and WoW, as the sustained transfer rate boost from RAID-0 does not overcome the added latency...few of the transfers are long enough to bring RAID-0's STR advantage to bear.
-------------------------
When running RAID-0, you are hurting the read ahead algorithms used by the disk...what the disk is reading ahead for and would ordinarily find and successfully retrieve for the system when running SLED may actually be on a different physical disk when running RAID-0.
-------------------------
But ... RAID-0 ... it's still faster
Increasing desktop storage performance is an art, not a science. Simply throwing more hardware at the problem may have impacts that you do not understand.
-------------------------
Part of the performance increase (if it does occur) in RAID-0 stems from simply the increased capacity of the logical volume, intensifying the effects of locality and reducing seek distances...an effect you can achieve by simply buying a drive that is twice as big. This is why running a single 250GB drive is better than running a pair of 80GB drives in RAID-0...you get the benefit without any of the drawbacks of RAID-0, for the same cost, while achieving more capacity.
-------------------------
Also, running RAID-0 increases the impact of rotational latency, as the spindles are not in sync. Rotational latency is in average. For a 10K drive, that average is 3ms. It takes 6ms for a platter to make a complete rotation at 10,000RPM. Sometimes the actuator will get in position for a request, then the data on the platter will pass underneath the heads fairly soon, say around 1ms. Other times, the heads may "just miss" the requested sector, and the drive may have to wait as 5ms or more for the platter to spin back around.
Let's look at the impact of rotational latency under a purely random load. If you are running SLED, 50% of the requests have a latency above 3ms, 50% below 3ms. Therefore it averages out. However, when you are running RAID-0, you have two spindles that are not in sync. Presuming that the requests "take advantage" of both drives, the entire point of running RAID-0, let's look what happens:
Drive 1 may have a < 3ms latency, and drive 2 may have a < 3ms latency. In that case, the rotational latency on that request is below average. If drive 1 takes > 3ms, but drive 2 takes < 3ms, the system may have to wait for both drives to complete their request, as the requested data may need to be processed to determine what to load next. In that case, the rotational latency is above average. Drive 1 may be < 3ms, but if drive 2 is > 3ms, you are once again faced with above average latency, and (obviously) if both drive 1 and drive 2 are > 3ms, then your rotational latency is above average.
Let's chart this up to make it clearer
SLED:
< 3ms: low
> 3ms: high
RAID-0:
< 3ms < 3ms low
> 3ms < 3ms high
< 3ms > 3ms high
> 3ms > 3ms high
When running SLED, 50% of the requests have an above average rotational latency, but when running RAID-0, 75% of the requests that require both drives to be serviced (isn't the goal of running RAID-0 to have two or more drives working on requests?) have above average rotational latency, thusly the performance decrease in the seek performance limited tests FarCry and WoW, as the sustained transfer rate boost from RAID-0 does not overcome the added latency...few of the transfers are long enough to bring RAID-0's STR advantage to bear.
-------------------------
When running RAID-0, you are hurting the read ahead algorithms used by the disk...what the disk is reading ahead for and would ordinarily find and successfully retrieve for the system when running SLED may actually be on a different physical disk when running RAID-0.
-------------------------
But ... RAID-0 ... it's still faster
Increasing desktop storage performance is an art, not a science. Simply throwing more hardware at the problem may have impacts that you do not understand.
If ALL you did was download, then obviously you shouldn't raid, no? You still don't understand the logic here. We're trying to find ONE real-world scenario where Raid would benefit the average user. Vague mumbling about "reading from blocks of memory and copying to disk" don't count.Lord of Shadows said:
So we've been working through the list of things users REALLY do. Gaming? No benefit. Office Apps? Newp. Downloading files? Not a chance. Working with video? Nyet. Copying large files? Sorry. What left? Feel free to chime in.
Sorry, chap, but you are utterly and completely wrong here. Thats the exact method drive manufacturers use to calculate MTBF rates...and the reason MTBF is separated from service life. With MTBF rates up to 40-50 years now, do you really think they test the drives that long? Of course not...they simply look at RMA rates for the first few months, divide by the number of drives sold, and spit out the result.
If a drive has a service life of 5 years and a MTBF of 300,000 hours, it means that in a collection of 300,000 drives, you get ONE FAILURE EACH HOUR. For the first five years that is...after that, your failure rate is expected to go up even higher.
So why were you claiming "copying large files" as a benefit of Raid? Oops.
No they wouldn't, because if they're doing transcoding or pretty much anything but a simple splice, they're cpu bound anyway, and Raid won't speed things up. And if they're just doing splices, cut, and whatnot, it'd be faster to put your video files on one drive, and your Windows swap file on a second, independent drive. That is, unless you're just splicing files smaller than the amount of available RAM you have.
So now we're down to the scenario of "splicing small video files, with no capture, transcoding, or othe conversion needing to be done". Why yes, I think RAID would benefit you there. How many users do that on a regular basis, eh?
No, not quite. First you buy a faster cpu. Then you buy more RAM and a faster video card. Then you buy a second video card and go SLI.EnderW said:
Once you're up to an FX-60 with 4 gigs of ultra highspeed RAM--- then and only then do you consider spending money on Raid. And only if you're regularly doing one of the very few things that benefit from it.
4GB RAM would hurt performance depending on what you were doing. And I already stated this same thing several posts up.masher said:
http://hardforum.com/showpost.php?p=1028836076&postcount=148
rogue_jedi
Supreme [H]ardness
- Joined
- Sep 6, 2002
- Messages
- 4,821
no, it really isn't. we are saying you shouldn't raid because it doesn't speed up your downloads and it costs more.Lord of Shadows said:
you mentioned getting a pair of SATA drives in raid 0 for $40 total; how big are they? i got my PATA drive (200gb diamondmax 9) for $40. there's hdtach results for it in the thread asking about the 6l200r0... give me a test you wish to compare your sata raid 0 to my pata disk with and i'll do it. i have another disk in the machine so i can time disk-to-disk copying.
EnderW: personally, i agree with you. because it does give some slight benefit, it is decent if you are going for the maximum in performance regardless of money. if you're trying to be cost-effective, there's better ways to improve performance.