Drive Selection for RAID

[EV700]

What are the most important factors when choosing a drive for RAID 0?

Is it interface performance? RAT? transfer performance? writing performance?

I am wanting to set up a RAID 0 for a gaming machine...

 

[ALANMAN]

I'd say something helpful here but every time I talk about RAID 0 someone goes on an Anti-RAID Crusade and tries to flame me to death.

Good luck in your quest for RAID.

 

[EV700]

ummk?

 

[choirbass]

the benefit of raid 0 for gaming is going to completely depend on the types of games you decide to play, and what youre expecting to see by going with raid 0. RTS games may see the most substantial improvement for level loading, depending largely on how the levels and maps are designed too, even (loading uncompressed bitmaps, for instance, can benefit largely from the improved STRs). current FPS games are more the other extreme, tending to show virtually no benefit from raid 0, for level and map loading. during gameplay however, you may see an improvement in gameplay smoothness, depending on how much the game needs to access your hdd. though usually having more than sufficient ram can help allieviate that problem.

overall, raid 0 provides the most substantial benefits for large file transfers (such as booting windows, for example), media editing, streaming, heavy disk accessing/thrashing (such as if youre running a file sharing server), synthetic benchmarks, and other similar uses.

it may actually be too, that faster random access times from higher rpms/or SSDs even provide more benefit for some games than having greatly increased STRs does.

 

[leo2kp]

I'd check out WD's enterprise disks as they have RAID specific timings and all that confusing jazz. 5 year warranty too.

 

[EV700]

choirbass: Very informative; I learned. But still, I want to know what factors are most important in choosing a hard drive for RAID 0.

leo2kp: WD is one brand of drive I'll never buy again; too many times have they clicked of death...

 

[choirbass]

really, its just a matter of making sure the 2 (or more) hdds are basically identical in specs, for the optimum performance... you can mix and match brands if you want, too

the capacity will be result of the smallest hdd times the number of hdds on the array (such as 1*250GB+2*320GB=750GB, any left over capacity is unusable). the performance improvement will also be at best a multiple of the slowest hdd in the array (make sure theyre all same speed). access times are never improved by going with raid 0, simply because you have more actuators, heads, and platters youre working with, compared to just a single hdd.

but really, any 2 or more hdds will work (they dont have to be designed for use in raid arrays), just make sure you have enough sata/pata ports available on the raid controller

 

[SomeJoe7777]

I would also recommend that to partially offset the risk of data loss from RAID 0 configurations that you use high reliability entry level enterprise hard drives that have high MTBF (mean time between failure) ratings and/or low AFR (annual failure rate) ratings.

This includes:

- Western Digital RE2 (1.2M hours MTBF)
- Western Digital Raptor (1.2M hours MTBF)
- Seagate Barracuda ES/ES.2 (1.2M hours MTBF, 0.73% AFR)
- Hitachi Ultrastar A7K1000 (1.2M hours MTBF, but targeted for use in a low duty-cycle environment).

 

[EV700]

I think I might just fold and get 2 74GB WD Raptors against my better judgment...

 

[choirbass]

tbh, raptors are among the available hdds that are less prone to failure. setting performance aside (since nearly everyone knows how fast they are), one of the things they come with is a 5 year warranty, so if by some chance something did happen to go wrong for whatever reason, you can always send it back for another new (most likely improved) replacement, up to half a decade later even (check their rma terms beforehand though, since the reseller needs to be supported by them for an rma under warranty to happen)

 

[chunkymonster]

The "I" in RAID stands for inexpensive. I've got my RAID0 using two 80GD WD's that I had left over from previous builds. The nice thing about RAID arrays is that they don't have to be made up of high end drives. As long as they are in good working order with no bad sectors any drives with the same specs will work just fine.

 

[leo2kp]

I've had one WD disk nearly fail on me but it was a Caviar 320GB drive. RAM'd it before it got too bad though. I'm overall very impressed with my WD Enterprise disks. I've had some Seagates go out on me tho

 

[enlightenment]

choirbass: its a myth that RAID improves STR only, please enlighten the tech community by mentioning it improves non-blocking IOps performance as well.

The most important factor in disks in a RAID is its interface, not the SATA/150 versus SATA/300 issue, but whether you used a PCI card to connect the disk or not. Onboard connectors provide the lowest latency, PCI-express follows closely, then comes PCI with a terribly huge latency, low bandwidth and high CPU utilization. So stay away from evil PCI and you'll be fine.

Also, don't buy those expensive "RAID drives" who have TLER and supposedly higher MTFS and 24/7 operation. It's just a method of trying to beat more money out of your pocket. After all, you get the same physical disk with just some firmware enhancements that really aren't so important.

 

[SomeJoe7777]

Also, don't buy those expensive "RAID drives" who have TLER and supposedly higher MTFS and 24/7 operation. It's just a method of trying to beat more money out of your pocket. After all, you get the same physical disk with just some firmware enhancements that really aren't so important.

Sorry guy, but I have to disagree with you there.

I'm not positive what's done to make a drive supposedly higher reliability (for example, what turns a Seagate Barracuda 7200.10, model ST3750640AS, into a Barracuda ES 7200.10, model ST3750640NS), but I know this: I work for a company that has 40 or so computers, several of them video editing workstations that have RAID arrays. Over the years, I've had over a dozen hard drives go bad for various reasons. Every one of the hard drives that went bad was a desktop model. Not one high-reliability drive has gone bad. I've used Maxtor Maxline IIs, II+'s, IIIs, Seagate Barracuda ES's, and WD Raptors. I have yet to have one of these fail.

Proof is in the pudding, my friend.

 

[enlightenment]

Sorry guy

Assumptions are fatal.

I'm not positive what's done to make a drive supposedly higher reliability (for example, what turns a Seagate Barracuda 7200.10, model ST3750640AS, into a Barracuda ES 7200.10, model ST3750640NS), but I know this: I work for a company that has 40 or so computers, several of them video editing workstations that have RAID arrays. Over the years, I've had over a dozen hard drives go bad for various reasons. Every one of the hard drives that went bad was a desktop model. Not one high-reliability drive has gone bad. I've used Maxtor Maxline IIs, II+'s, IIIs, Seagate Barracuda ES's, and WD Raptors. I have yet to have one of these fail.

Proof is in the pudding, my friend.

You cannot draw conclusions out of 40 disks. If you want proof, you should consider 20 million or so desktop models. Even the statistics Google has gathered are not really representative. Maybe the disks that failed more were inserted in specific models who have less cooling, were more heavily used, had bigger temperature range, more up and down cycles etc.

The "RAID edition" types (MaxLine, Barracuda ES, WD Raid Edition) are physicically the same as their desktop counterparts, they come of the same factory line. The difference is in testing, perhaps picking the 'best samples' and other firmware. I've had Maxline drives too (12 of them, to be exact) and 3 of them failed. Does that make Maxline's unreliable? I do not draw such conclusions from such a limited number of disks. Just read the forums, everyone hates a HDD brand because he or she had x disks fail of that brand and NONE of the brand he or she is currently using has failed yet. So what? That really doesn't say anything.

That said, i can understand that you have good experiences with the more expensive 'RAID' drives and for that reason alone you want to pick them. That's your good choice ofcourse and for the added warranty alone that might make sense. Personally i don't share your opinion and think its just way for manufacturers to 'rebadge' the disks to sell them for more profit.

 

[russki]

choirbass: its a myth that RAID improves STR only, please enlighten the tech community by mentioning it improves non-blocking IOps performance as well.

Your point is well taken, but I have to agree with the greater point choirbass is making and that is that RAID0 is practically useless (and in general a bad idea) for a gaming desktop. And besides, I am pretty sure you are talking on a level of understanding well above that of the OP or, for that matter, the vast majority of people seeking advice on these forums.

 

[ALANMAN]

Promised myself I wouldn't post about raid, but here I am. I'd recommend the WD RE2 Drives, like many others, I have 2 of the 320 GB models in RAID 0, they work great.

Don't know if this is true, but I heard that the RE2 drives have better bearings and higher quality parts. Anyone know if this is true?

 

[enlightenment]

Your point is well taken, but I have to agree with the greater point choirbass is making and that is that RAID0 is practically useless (and in general a bad idea) for a gaming desktop. And besides, I am pretty sure you are talking on a level of understanding well above that of the OP or, for that matter, the vast majority of people seeking advice on these forums.

I agree with you that RAID0 for a gaming desktop doesn't provide much additional performance, maybe 10% in loading scenes if you're lucky. But this is also caused by stripe misalignment because Microsoft never considered the use of NTFS on a RAID. If they had they would have used a partition offset of 1MB, so that the filesystem and stripes were aligned properly. Without it, you would loose the parallellism potential of RAID0 in random reading.

The fact that the TS might not understand the theory about RAID and storage in general does not mean it is irrelevant to him; since he is affected by it. After all he's coming here for advice, so its logical to be confronted by things that are new to him. Aside form additional performance, RAID0 does have the benefit of one big filesystem as opposed to seperate disks. And for other desktop tasks RAID0 ofcourse does provide a nice benefit.

 

[russki]

The fact that the TS might not understand the theory about RAID and storage in general does not mean it is irrelevant to him; since he is affected by it. After all he's coming here for advice, so its logical to be confronted by things that are new to him. Aside form additional performance, RAID0 does have the benefit of one big filesystem as opposed to seperate disks. And for other desktop tasks RAID0 ofcourse does provide a nice benefit.

No, not irrelevant. Just maybe a bit over the head. For example, the points that you make are really good, but maybe you could expand on the reasons why. We really need more well informed posters like you around here. But it is also true that we can learn more from understanding the reasons and not just the consequences.

 

[SomeJoe7777]

Assumptions are fatal.

I wasn't using the word "guy" in a gender sense, but rather just as a greeting to a person on the forum. Much the same way that people informally say "dude". I don't believe gender should matter on the forum, anyway.

You cannot draw conclusions out of 40 disks. If you want proof, you should consider 20 million or so desktop models. Even the statistics Google has gathered are not really representative. Maybe the disks that failed more were inserted in specific models who have less cooling, were more heavily used, had bigger temperature range, more up and down cycles etc.

The "RAID edition" types (MaxLine, Barracuda ES, WD Raid Edition) are physicically the same as their desktop counterparts, they come of the same factory line. The difference is in testing, perhaps picking the 'best samples' and other firmware. I've had Maxline drives too (12 of them, to be exact) and 3 of them failed. Does that make Maxline's unreliable? I do not draw such conclusions from such a limited number of disks. Just read the forums, everyone hates a HDD brand because he or she had x disks fail of that brand and NONE of the brand he or she is currently using has failed yet. So what? That really doesn't say anything.

That said, i can understand that you have good experiences with the more expensive 'RAID' drives and for that reason alone you want to pick them. That's your good choice ofcourse and for the added warranty alone that might make sense. Personally i don't share your opinion and think its just way for manufacturers to 'rebadge' the disks to sell them for more profit.

Fair enough. In all likelihood, you're correct, that the high reliability series drives are simply hand-picked samples that showed more MTBF during the abbreviated testing. I have no problem with this -- Intel and AMD take choice processor cores from the center of the wafer and sell them as certified for higher clock speeds, while those from the edge of the wafer are sold as lower speed models. The testing reveals what the part can take.

If I'm paying more for exactly the same drive, but testing showed that for whatever reason, one particular batch tests notably better than another, that's reason enough to pay slightly more for those drives. Assuming, of course, that the manufacturer isn't outright lying to us and has actually done testing that statistically shows the higher reliability.

Who knows why that batch tested better? Maybe the air handlers' filters just got changed that day. Maybe the plant temperature was 68F instead of 72F during assembly. Maybe Steve was on the head assembly line today instead of John. The point is that no matter what, there are some factors that will always be variable that you can't control in an assembly process, and in high tolerance applications like this, can make a substantial difference in the final outcome.

 

[enlightenment]

I wasn't using the word "guy" in a gender sense, but rather just as a greeting to a person on the forum. Much the same way that people informally say "dude". I don't believe gender should matter on the forum, anyway.

Hehe ok fair enough.

If I'm paying more for exactly the same drive, but testing showed that for whatever reason, one particular batch tests notably better than another, that's reason enough to pay slightly more for those drives. Assuming, of course, that the manufacturer isn't outright lying to us and has actually done testing that statistically shows the higher reliability.

Well this is the key point i guess. I'm not even sure they are 'picked' this way, but it is a possibility. Maybe the manufacturers are simply rebadging like 5% of all casual drives as RAID drives with added warranty and different firmware and that's it. If that is the case, then i would think the drives offer no advantage in reliability. Maybe the firmware could change something, but it cannot change the hardware capabilities of the drive.

Personally i think they simply test the expensive drives more to ensure they are really good when they leave the factory, but that you can't simply evaluate how 'good' a sample really is. So its not like oh this disk scores 98% so its gonna be a MaxLine drive, this one scores only 68% so its gonna be a consumer drive. It is possible but i doubt it. I think you would need to disassemble and inspect the disk under microscope or the like to really evaluate it and say something about the quality of this particular sample. However, due to the extra testing there are no really bad ones leaving the factory labeled as MaxLine/RE/ES drives, which otherwise would be sent back as a DoA (Dead on Arrival).

 

[enlightenment]

No, not irrelevant. Just maybe a bit over the head. For example, the points that you make are really good, but maybe you could expand on the reasons why. We really need more well informed posters like you around here. But it is also true that we can learn more from understanding the reasons and not just the consequences.

Okay, i'll try to explain it a bit. Storage can be pretty complicated though, surprisingly, many reviewsites make huge mistakes. That's why i decided to write articles about storage, but i have not finished them yet.

Stripe misalignment
In a striping RAID scheme the data is stored non-contigious on the disks. This means that to read 1MB you cannot read all that from a single disk in a RAID0 configuration. You can with RAID1, because that scheme uses no striping. RAID1 is unaffected by stripe misalignments.

So what is a misalignment? Well, your filesystem (FAT/NTFS/EXT3/UFS) has to begin somewhere. You would think it starts at the beginning of the volume (a single disk or RAID volume), but it doesn't. It leaves space in front of the partition to store the DOS partition table ("fdisk") and to store the master boot sector for BIOS boot support. Each sector is 512 bytes large. Normally the first partition starts at an 'offset' of 63 sectors. Let's see this visually:

[fixed]| 63 sectors for partition table | 40GB NTFS partition | 20GB other partition |[/fixed]

In a single disk configuration this is no problem at all, however in a striping (RAID0/3/4/5/6) configuration there are performance difficulties. If your stripesize is 64KB and you want to read 64KB then ideally this means you have to access just one disk. A subsequent 64KB request can be processed by another disk member, this is the parallellism effect RAID0 gains its performance from. If there is a misalignment between stripe blocks and the filesystem offset, then a file on the filesystem will nog start on the beginning of a stripeblock. Meaning: a file will not start at offset 0KB of a 64KB stripeblock but at 31.5KB for example. If you then read 64KB, you have to read from 2 disks instead of 1!

[fixed]
| 64KB stripe on disk1 | 64KB stripe on disk2 | 64KB stripe op disk3 |
| unused | 64KB FILE ON FILESYSTEM | ~~another 64KB block~~ | [/fixed]

As you can see, the blocks on the filesystem do not start at the same point as a stripe block. This is called a misalignment. To read the first 64KB block, you need to read both from disk1 and disk2. That means both disks are busy with 1 request and that both are not available for other I/O requests. Ideally you want one I/O request to be handeled by 1 disk, so that the other disks are able to process other requests. With a stripe misalignment this effect is hindered and sometimes effectively negated. A solution is to use a 1MB offset for your filesystem, something Microsoft should have done if they had any clue:

[fixed]
| 64KB stripe on disk1 | 64KB stripe on disk2 | 64KB stripe op disk3 |
| ~~~~unused space~~~~ | 64KB FILE ON FILESYS | ~another 64KB block~ | [/fixed]

Here you have a 64KB offset at the beginning of the filesystem; now they are perfectly aligned! However you still have a misalignment if you use a stripesize of 128KB or higher. So probably an offset of 1MB should be used. Then it would consume 8x 128KB stripes or 16x 64KB stripes as 'unused' (though they hold the boot code and partition table). Another solution is to use Dangerously Dedicated mode, in which there are no partitions. Then the filesystem starts at offset 0. Windows does not support this mode to my knowledge.

 

[russki]

Thanks for the post. Yeah, storage is complicated, and it is generally underappreciated just how much, mostly because most people learn about storage on a high level, but it really is the most complicated on the low level - file system structure, and layout on the drive. I for one, do not have a clue how NTFS is laid out, and I don't think documentation is readily available. At the time I learned any of it, it was FAT (yeah, not FAT32). And there are other issues with the structure, as in the balance between searchability and sequential access, and more recently parallelism, etc. File systems is quite a complicated subject.

Also, I had to think about your 1Mb offset point, 'cause the stripe size is smaller. But, as you said, it does solve the problem.

It's good to see you explain it like this.

 

[EV700]

So random access times is the answer?

 

[EV700]

/bump