For those thinking about running RAID

[Pariah]

Tech Report has posted a very indepth RAID performance test of multiple different controllers. For those too lazy to read the entire article, here are the highlights:

Read times don't scale quite as dramatically as file creation times, but there's still a clear advantage to RAID.

That's good.

File Copy Test - Copy time

And we get another glorious example of RAID scaling...Otherwise, all the systems see faster performance moving from single drives to RAID 1 and 0.

That's good too.

Unfortunately, Business Winstone scores don't scale much at all. The performance benefit moving from a single drive to RAID 1 or RAID 0 is negligible at best.

This isn't good.

We see a little more scaling action in the Multimedia Content Creation Winstone, particularly moving to RAID 0, but the small performance boost probably isn't enough to justify the cost of a second hard drive.

Not much better here.

RAID doesn't offer much of a performance benefit over single-drive configurations in our DivX encoding test. Sure you can pick up a fraction of a frame per second by adding a second drive, but really, that's about it.

Wouldn't expect any boost here. CPU limited.

With the exception of the SiS964, which boots slowly with only a single drive, RAID doesn't do much to speed up boot times.

Same deal. Nothing worthwhile.

Perhaps our level load times will be more interesting...

Sorry. Nope.

Notice a pattern here?

Again, going to RAID doesn't speed things up a whole lot. The SiS964 and nForce3 250Gb both flirt with a one second slow-down moving from a single drive to RAID 1, but level load times are otherwise pretty consistent in the Unreal Tournament 2004 demo

Final thoughts?

Of course, I would be remiss not to point out how little performance impact different RAID levels had in our application benchmarks and stopwatch tests. Although multi-user and synthetic disk subsystem tests like IOMeter, HD Tach, and ATTO show clear differences between the performance of each RAID implementation and array configuration, the performance benefit in more real world applications is significantly less pronounced. Power users who regularly stress IO performance will no doubt notice a performance benefit moving from a single drive to striped array, but for less demanding users, the redundancy benefits (with no performance loss) associated with a mirrored array may be a better justification for adding a second drive.

Not a very glowing view of RAID. But hey, if you want to spend twice as much to add a second drive, or more if you need a RAID controller too, while increasing your chances of data loss just so you can think you look cool by saying you run RAID, you're certainly free to do so.

edit:

duh, forgot to post a link to the article:

http://tech-report.com/reviews/2004q2/chipset-raid/index.x?pg=1

 

[Nebor]

Not a very glowing view of RAID. But hey, if you want to spend twice as much to add a second drive, or more if you need a RAID controller too, while increasing your chances of data loss just so you can think you look cool by saying you run RAID, you're certainly free to do so.

That's me. My home leisure computer has games, chat logs and favorites on it. Getting to say "RAID, biatch" is worth that stuff.

 

[Markfw]

This only applies to IDE/SATA on-board raid I am sure. Caching controllers and SCSI are a whole other animal.

 

[3chordcharlie]

I can't se much benefit to raid-0, and don't have any data I consider valuable enough to justify raid-1 (though many people do have data they would hate to lose). The way I see it, for a performance-driven user, raid-5 is probably the way to go, since it provides some security. I understand raid-5 doesn't provide huge performance gains either, but it makes better use of disk space while providing redundancy.

EDIT: speaking about IDE RAID, of course

 

[Pariah]

Originally posted by: Markfw900
This only applies to IDE/SATA on-board raid I am sure. Caching controllers and SCSI are a whole other animal.

Caching is not going to improve load times, boot times, or typical application performance for anything using RAID 0 or 1 no matter what interface or implementation you are looking at.

 

[nitromullet]

nice link to the article btw...

 

[WobbleWobble]

There are many motherboards with built-in RAID capability. If you have a 120G drive and need more space, why not just RAID them? In the case of RAID-0, it provides you with a slight boost. I have a lot more faith in the reliability of HDDs than some here. When I did run RAID-0, it definately felt like map loading times were a lot quicker. But generally, didn't feel anything.

I don't ever plan on running RAID-1 because I think I'm more likely to screw up by accidentally deleting a file or getting hit by a virus than having a drive fail. A regular backup on DVD or something is a much better idea.

That all said, I have ran RAID a few years ago but I no longer do. But if I do decide to upgrade my HDD space, I could very easily just drop in an identical drive and get some "free" performance, even though it's minimal.

 

[sharkeeper]

Caching is not going to improve load times, boot times, or typical application performance for anything using RAID 0 or 1 no matter what interface or implementation you are looking at.

Ugh, not quite. Properly implemented caching schemes help tremendously reducing mechanical disk activity that is clearly visible.

The difference using just one disk (from a simple non intelligent HBA) to high end controller is amazing. Using 1-2GB of write back cache with dynamic flush timings is very beneficial.

Cheers!

 

[Pariah]

Leave it to shuttle to post a complete nonsense setup that has no useful context to a setup any of us would actually consider using. Even so, there is still no reason 90TB of brilliant caching will make an ounce of difference in boot performance or Unreal Tournament loading. Cache can't make the disc read any faster than it is capable of.

 

[Jeff7181]

Originally posted by: Pariah
Leave it to shuttle to post a complete nonsense setup that has no useful context to a setup any of us would actually consider using. Even so, there is still no reason 90TB of brilliant caching will make an ounce of difference in boot performance or Unreal Tournament loading. Cache can't make the disc read any faster than it is capable of.

And in a perfect world where the drive is operating at 100% efficiency and is only limited by spindle speed, yes, you're right. But here in the real world, things are not 100% efficient and hard drives certainly are not ONLY being limited by spindle speed.

*EDIT* Imagine for a bit...

Hard drives average about 50 MB/s transfer rates. Lets say you have a 5 GB cache. You open Far Cry. The ENTIRE Far Cry directory is about 3 GB. So... if it caches the whole Far Cry directory, it'll take 60 seconds at 50 MB/s. So you have an intial load time of 60 seconds. Then after it's in the cache, it's only limited by the speed of the bus that the drive is on, which is 150 MB/s for SATA. So you move onto the next map which is, lets say... 150 MB... so it takes 1 second for the next map to load.

That's assuming the cache is located on the drive or controller that resides on the SATA bus. Imagine if you had 4 GB of system RAM, and some of that could be used to cache hard drive info... you'd spend 60 seconds loading, then after that everything would be almost instant... with single channel PC3200 RAM delivering up to 3.2 GB/s... or up to 6.4 GB/s with dual channel PC3200 RAM.

That's hardly practical though.

What would be more practical, and I can't figure out why no manufacturers are even attempting it, is drives with "dual" heads positioned 180 degrees from eachother. Either head can be reading or writing at the same time... so transfer rates would be doubled... rotational latency would be halved because either head could begin reading as soon as the data passes under it... seek times would be nearly halved. It would be sort of like RAID 0 in a single drive, only better. If they could do that, then they could make drives with 3 heads positioned 120 degrees from eachother... and 4 heads 90 degrees from eachother... you get the point. Screw 8 MB caches... with the size of solid state storage devices now days you could easily put 128, 256, 512, or even 1 GB cache in a hard drive. The only problem with that would be with power failures... lots of unwritten info could be lost if the power shuts off and it never gets written to the disk.

Ok... I'm done, lol.

 

[nitromullet]

thx for the link...

I didn't bother to read the rest of what you term a "very indepth RAID performance test" after reading this...

...Intel, NVIDIA, SiS, and VIA don't agree on a common default stripe size. Intel suggests a 128KB strip size; NVIDIA prefers 32KB; and SiS and VIA default to 64KB. I trust that each company has a firm enough grasp of the unique performance characteristics of its RAID controller to recommend the best stripe size for its chipsets, so we'll be using default stripe sizes throughout....

Actually sounds more like a pretty lazy approach to testing and a bad assumption to make, if you ask me. It is pretty widely known and accepted that different stripe sizes can have a tremendous affect on the performance of the array in different applications. I don't doubt the results they got from their testing, nor would I encourage anyone to build themselves a RAID 0 array because it offers "blistering" performance for the money. However, I don't think their testing methodology could be described as "very indepth".

 

[Pariah]

Originally posted by: Jeff7181

Originally posted by: Pariah
Leave it to shuttle to post a complete nonsense setup that has no useful context to a setup any of us would actually consider using. Even so, there is still no reason 90TB of brilliant caching will make an ounce of difference in boot performance or Unreal Tournament loading. Cache can't make the disc read any faster than it is capable of.

And in a perfect world where the drive is operating at 100% efficiency and is only limited by spindle speed, yes, you're right. But here in the real world, things are not 100% efficient and hard drives certainly are not ONLY being limited by spindle speed.

Let me modify my original statement as you are correct. What I should have said was, additional caching to what we already have will not make any worthwhile increase in performance. Adding cache to a storage setup that has no cache will improve performance, but if properly configured, you'll rapidly reach a point of diminishing returns by adding more. Going beyond 8MB isn't going to yield anything worth while for a home user. Even SCSI doesn't go beyond 8MB anymore. It makes a whole lot more sense to load up on system RAM and let the system cache disk reads than it does to put the cache on the drive itself where it can only burst to the system at the limit of the interface.

 

[StraightPipe]

"RAID, biatch"

no, wait, my RAID is broken

maybe next week

 

[Jeff7181]

Oh... and I hate when people use "RAID" as a verb :roll: "I RAIDED two Raptors dawg! It's mad fast yo!"

 

[sisooktom]

Agreed that RAID is somewhat cool but almost totally useless for most users. You get a slight improvement in performance for double the risk of data loss, no thanks.

 

[Pariah]

Originally posted by: nitromullet
thx for the link...

I didn't bother to read the rest of what you term a "very indepth RAID performance test" after reading this...

...Intel, NVIDIA, SiS, and VIA don't agree on a common default stripe size. Intel suggests a 128KB strip size; NVIDIA prefers 32KB; and SiS and VIA default to 64KB. I trust that each company has a firm enough grasp of the unique performance characteristics of its RAID controller to recommend the best stripe size for its chipsets, so we'll be using default stripe sizes throughout....

Actually sounds more like a pretty lazy approach to testing and a bad assumption to make, if you ask me. It is pretty widely known and accepted that different stripe sizes can have a tremendous affect on the performance of the array in different applications. I don't doubt the results they got from their testing, nor would I encourage anyone to build themselves a RAID 0 array because it offers "blistering" performance for the money. However, I don't think their testing methodology could be described as "very indepth".

Makes perfect sense to me what they did. All the implementations that tested are all targetted towards home users. Why wouldn't these companies choose the default size that was best for their implementation? The test suite TechReport ran was also clearly targetting applications and usage patterns that home users would typically use, so there was no point in changing the cluster size from default.

 

[yhelothar]

I just took the raid off my two 160GB maxtors after a hard drive corruption.
I have noticed a DRAMATIC change in performance. Mostly because I copy a lot of large files and do a lot of file compression/decompression. With RAID, it takes less than half the time! Especially when I'm multitasking while decomressing, it takes 1/4th the time with RAID.. I don't know exactly why. With raid, it takes me 30 seconds to decompress a 500MB file, while without raid, it takes just over 2 minutes!
If i decompress without multitasking, I get 20 seconds with RAID, and 40 seconds without RAID..

 

[Markfw]

Originally posted by: Pariah

Originally posted by: Jeff7181

Originally posted by: Pariah
Leave it to shuttle to post a complete nonsense setup that has no useful context to a setup any of us would actually consider using. Even so, there is still no reason 90TB of brilliant caching will make an ounce of difference in boot performance or Unreal Tournament loading. Cache can't make the disc read any faster than it is capable of.

And in a perfect world where the drive is operating at 100% efficiency and is only limited by spindle speed, yes, you're right. But here in the real world, things are not 100% efficient and hard drives certainly are not ONLY being limited by spindle speed.

Let me modify my original statement as you are correct. What I should have said was, additional caching to what we already have will not make any worthwhile increase in performance. Adding cache to a storage setup that has no cache will improve performance, but if properly configured, you'll rapidly reach a point of diminishing returns by adding more. Going beyond 8MB isn't going to yield anything worth while for a home user. Even SCSI doesn't go beyond 8MB anymore. It makes a whole lot more sense to load up on system RAM and let the system cache disk reads than it does to put the cache on the drive itself where it can only burst to the system at the limit of the interface.

OK, I have SCSI and I have 128 meg cache. So where does that leave your statement ? And load time CAN improve with 2-5 (in my case) disks retrieving information at the same time. When I get time, I have to benchmark my two systems load times on Far Cry levels. I will reply back when I get time, but I can tell you from experience, my SCSI raid array wipes the floor with my other system, and an AthlonXP at 2200 mhz is not that much slower than my Athlon64 3000@2130. (cpu usage for loading times of data, other things are faster)

 

[dullard]

Isn't that basically the results and conclusions of every other ATA RAID review? Anandtech did a very thorough one a while ago that said basically the same exact thing. Link to relavant Anandtech page. 3%-13% faster in real world test when running at optimum RAID conditions.

 

[sharkeeper]

Cache can't make the disc read any faster than it is capable of.

This isn't about the physical disk characteristics. It's about the storage subsystem. It's quite obvious you have no experience with high end storage deployment and use. :|

I never implied that cache performs miracles. It has a purpose which few people here actually understand.

Cheers!

 

[Ryoga]

Ubiqutous RAID link

 

[Pariah]

Originally posted by: dullard
Isn't that basically the results and conclusions of every other ATA RAID review? Anandtech did a very thorough one a while ago that said basically the same exact thing. Link to relavant Anandtech page. 3%-13% faster in real world test when running at optimum RAID conditions.

Pretty much. And you would think this concept would eventually catch on with users, but I guess it's not to be. How many hours do you think it will be before someone asks on this board how much of a performance gain they will get by buying a second drive and creating a RAID array or whether or not they should use RAID, or whether they should buy 2 36GB Raptors, or one 74GB?

 

[EeyoreX]

There are many motherboards with built-in RAID capability. If you have a 120G drive and need more space, why not just RAID them? In the case of RAID-0, it provides you with a slight boost. I have a lot more faith in the reliability of HDDs than some here. When I did run RAID-0, it definately felt like map loading times were a lot quicker. But generally, didn't feel anything.

I have seen this many times too. The placebo effect often is greater than any real benefit.

I don't ever plan on running RAID-1 because I think I'm more likely to screw up by accidentally deleting a file or getting hit by a virus than having a drive fail. A regular backup on DVD or something is a much better idea.

I agree, that using RAID1 as a sole backup is a bad idea.

That all said, I have ran RAID a few years ago but I no longer do. But if I do decide to upgrade my HDD space, I could very easily just drop in an identical drive and get some "free" performance, even though it's minimal.

The effort required to add more space via creating a RAID array is not worth it. If I want to "upgrade my HDD space" I'll plug in another drive and be done with it. Why go through the effort of either backup and restore or reinstall from scratch just to add some space and get a marginal, at best, boost?

RAID of course has it's place and its uses. The general purpose desktop is not one of those places.

\Dan

 

[filmore crashcart]

I, for one, thank you for posting this. You have saved me time and money on something I was going back and forth about because a lot of people seemed to be pushing raid-0 as the way to go to increase performance. Having just one, nice size hard drive is so much neater, and cheaper, in setting up.

 

[robcy]

While RAID is useless for the majority of users, there are cases where its benefits are quite usable. I have an 80gb 8mb cache 7000rpm (51MB/s) primary drive for the OS, and the Office 2003 stuff. I also have two 40gb 2mb 5400rpm drives on a RAID 0 that are used for games. Now, individually these drives do about 33MB/s but in RAID 0 they do 61MB/s this allows IL2 FB to load in 8 to 10 seconds compared to 20 to 22 seconds when the drives where seperate or 15 to 16 seconds when I used the 80gb 8mb HD as the game drive. All of my other flight sims have similar speed ups from the RAID 0 setup. This alows me to get pretty good performance from cheap drives, and a $13 RAID card, and if one burns, so what, I still have the CDs, and I will just load them again.