RAID 0,5, or 6 for gaming?

[simplyput]

Ok, I understand that RAID 0 is the current way to go for fast loading times in games... but what about RAID 5 or 6? I currently have a DFI lan party mobo with an integrated RAID 5 controller, and I have no idea if it's worth hooking up. I guess my main question is this:

Is RAID 5 or 6 using 3 hard drives any faster than RAID 0 using 2 of those same hard drives? (read speed)

 

[222Doc]

i use two raptors in raid O and i load almost first every time in BF2 sometimes i loads so fast that i can cap a flag before the next guy even gets to spawn i think a other point is sys mem 2 gigs so maps stay in mem, not sure but i noticed a great leap when i swaped it to 2gigs

 

[hubbardt]

RAID5 or 6 provides you redundancy so that if a disk fails you don't lose all your data. The downside is that if you RAID5 3 disks you lose one disk as storage capacity and each write cycle has to write additional parity information.

Stick with RAID0 ....

 

[mr_fnord]

RAID 5 can allow reads from all spindles simultaneously, so 3 spindles in RAID 5 will have better performance of two in RAID 0, as well as redundancy. This assumes high quality HW RAID implementation, SW or cheapo implementation won't get you the full performance gains.

During small random reads there is a chance that you will not get your performance increase, if you are reading random individual blocks you will read from all spindles -1, which would put you on par with RAID 0 with two drives. Generally you won't notice this condition, if block sizes are 64K the read from 2 spindles would be so fast you wouldn't notice the decrease in speed.

RAID 6 is just RAID 5 with extra parity, the theory being it can withstand multiple simultaneous failures. Can't do RAID 6 with 3 drives, and you obviously aren't too concerned with data integrity.

Depending on the brand and quality of your RAID controller, RAID 5 should be noticably faster than RAID 0. You also get redundancy, a HD failure won't phase you. But, cheap implementations might actually be slower due to added overhead, and you will have to buy an additional drive.

 

[simplyput]

RAID 5 can allow reads from all spindles simultaneously, so 3 spindles in RAID 5 will have better performance of two in RAID 0, as well as redundancy. This assumes high quality HW RAID implementation, SW or cheapo implementation won't get you the full performance gains.

During small random reads there is a chance that you will not get your performance increase, if you are reading random individual blocks you will read from all spindles -1, which would put you on par with RAID 0 with two drives. Generally you won't notice this condition, if block sizes are 64K the read from 2 spindles would be so fast you wouldn't notice the decrease in speed.

RAID 6 is just RAID 5 with extra parity, the theory being it can withstand multiple simultaneous failures. Can't do RAID 6 with 3 drives, and you obviously aren't too concerned with data integrity.

Depending on the brand and quality of your RAID controller, RAID 5 should be noticably faster than RAID 0. You also get redundancy, a HD failure won't phase you. But, cheap implementations might actually be slower due to added overhead, and you will have to buy an additional drive.

Thank you, that is exactly the information I was looking for... It did, however, spawn 2 more questions in my mind:

1) How can I identify a good RAID 5 controller?
(Is the one on my DFI lanparty UT good enough?)

2) Can I get a performance bottleneck if I don't go with SATA2?
(or does each drive just use it's own bandwidth?)

Currently I'm using a single 74Gb Raptor, and from what I understand, in RAID 0 they can flood the SATA bandwidth on large data transfers.

 

[enewmen]

Ya
RAID 5 is a big investment.
$200 for the controller PLUS 3+ drives.

I will personally get ONE good Raptor drive for the OS, Games, etc.
Then add the RAID 5 for just storage.
If the Raptor crashes, just re-install everything from CDs since all the data is still safe in the RAID 5 array.

 

[svyerkgeniiy]

No no no. You're so half-right that you're wrong.

RAID 5 can allow reads from all spindles simultaneously

So does RAID 0.

3 spindles in RAID 5 will have better performance of two in RAID 0

Not true. At most you get the same performance since only 2 spindles are used to hold data while the third holds a error-correcting check. You generally get slightly worse on reads because of the overhead of the calculations for the third drive. Writes to a RAID 5 are almost always slower since you have to generate the error-correction, so it depends on your controller. Add-in controllers are faster than built-in motherboard controllers.

Depending on the brand and quality of your RAID controller, RAID 5 should be noticably faster than RAID 0

No it won't. If you have RAID 0 all drives are used for data so you get about the number of drives times the speed of a single drive; but for RAID 5 you get the number of drives minus one times the speed of a single drive, so reads are slower for the same number of drives. Also writes are significantly slower, depending on the controller because the controller has to create the extra drive's correction codes.

If gaming speed is your only concern, stick with RAID 0. However, multiple drives mean that one drive failure out of the set will cause the whole array to fail. For example, if you have a 1% chance of failure per drive, four drives in RAID 0 will give you 99%^4 chance of no failure = almost 4% chance of failure instead of 1%. However, with four drives in RAID 5 you need two drives to fail, meaning that your chance of failure means about 1/100 of 1% (= 0.0001) of simultaneous failure times 6 ways to fail means 99.99%^6 or about a .06% chance of loss of the whole array. .06% vs 4% means a factor of 67 difference in reliability in this case for RAID 5 vs RAID 0.

So I'd say, if you have only 2 drives in RAID 0, and you use the RAID only for non-critical data, your drive setup is probably okay. If you're using more drives, consider RAID 5. You won't get the maximum speed, but you won't lose everything to a single crashed hard drive.

--dv

 

[svyerkgeniiy]

1) How can I identify a good RAID 5 controller?
(Is the one on my DFI lanparty UT good enough?)

Look for reviews on the net. Tom's recently did a review. In general add-in cards are better than built-in motherboard controllers; Areca seems to be testing out to be the best. But the tend to be pricey, from $200 for a 4-drive board and on up from there. Areca's 4-drive controller is about $350 direct.

You on board controller is probably okay for RAID 0, but the onboard ones tend to lose out more for more complex forms of RAID (like 5).

2) Can I get a performance bottleneck if I don't go with SATA2?
(or does each drive just use it's own bandwidth

Depends on the controller. PCI controllers are limited by bus bandwidth; to get the full benefit you really need PCI-X 64-bit connectors at 66 MHz for large arrays. PCI-e are better but require newer motherboards and are harder to find. On board controllers depend on how they transfer the data around, though I'm unsure what internal connections they use. SATA2 vs SATA shouldn't be a problem as long as the drives go through the controller; that's where the data rate will be limited. Single drives have a hard time flooding even a SATA1 bandwidth, but if that's all the controller uses for output, then maybe a RAID 0 array will flood the output.

--dv[/quote]

 

[psyno]

2) Can I get a performance bottleneck if I don't go with SATA2?
(or does each drive just use it's own bandwidth?)

So there are two answers to this, depending on what exactly you meant.

The first is no, not really, in practical sense anyway. Each SATA device is given a dedicated link, and so they aren't sharing the 1.5 Gbps. So, with current drives, the link will rarely be saturated for a significant period.

The other answer is, yes, the drives may saturate other buses as svyerkgeniiy indicated, but the situation is no different whether you're using SATA 1.5 Gbps or SATA 3.0 Gbps, since the bottleneck isn't between the controller and the disk. So really, no.

(edit: typo)

 

[simplyput]

2) Can I get a performance bottleneck if I don't go with SATA2?
(or does each drive just use it's own bandwidth

Depends on the controller. PCI controllers are limited by bus bandwidth; to get the full benefit you really need PCI-X 64-bit connectors at 66 MHz for large arrays. PCI-e are better but require newer motherboards and are harder to find. On board controllers depend on how they transfer the data around, though I'm unsure what internal connections they use. SATA2 vs SATA shouldn't be a problem as long as the drives go through the controller; that's where the data rate will be limited. Single drives have a hard time flooding even a SATA1 bandwidth, but if that's all the controller uses for output, then maybe a RAID 0 array will flood the output.

--dv

Ok, let me try this I have a DFI Lanparty UT nF4 SLi-D mobo. It has integrated SATA and SATA2 controllers, as well as integrated RAID
0, 1, 0+1, and 5.

Now the main thing I'm looking for is read performance so I can eliminate long loading times. Like I said, this is going to be used mainly for gaming. I'm not too worried about HD failure (should I be, does it happen often enough?). So the questions I have now are:

1) Where is the first bottleneck to read time going to occur?
-SATA bandwidth? -the chipset? -RAID controller? (I have no idea)

2) Is there a RAID 4 or some other form of RAID that allows for redundancy and multiple spindle reads without any error check? (A gaming RAID, if you will)

3) How big of a difference will I see in the read speed of a software RAID 5 controller vs a hardware one?

 

[psyno]

1) Where is the first bottleneck to read time going to occur?
-SATA bandwidth? -the chipset? -RAID controller? (I have no idea)

The first and biggest bottleneck is from the physical medium to the electronic state.

2) Is there a RAID 4 or some other form of RAID that allows for redundancy and multiple spindle reads without any error check? (A gaming RAID, if you will)

Every RAID mode does multiple spindle reads, and every RAID mode but RAID 1 stripes data in some fashion. Parity checks are generally done on writes, not reads. If you're not interested in redundancy go RAID 0. If you are interested in redundancy but you're worried about slow read performance because of the error checking, then it's something you need to worry about at the controller level and not at the RAID level (pun more or less intended). This isn't to say that different RAID levels don't have different read performances, just that from the theoretical properties of RAID, it's not the redundancy check that kills you on the read, regardless of the level.

 

[hubbardt]

I think we are getting a bit too complicated here !

RAID0 is for power gamers who want fast loading times
RAID1 is a waste for home users tbh. too much redundancy
RAID5 is too expensive for home use

RAID1 and RAID5 provide redundancy incase of disk failure. If you are that concerned about drive failure, buy a DVD/tape backup solution

nuff said

8)

 

[lostny11]

2 SSD's in raid0, untouchable!!!!!!!!

 

[rook512]

3 spindles in RAID 5 will have better performance of two in RAID 0

Not true. At most you get the same performance since only 2 spindles are used to hold data while the third holds a error-correcting check. You generally get slightly worse on reads because of the overhead of the calculations for the third drive. Writes to a RAID 5 are almost always slower since you have to generate the error-correction, so it depends on your controller. Add-in controllers are faster than built-in motherboard controllers.

The parity portions are spread out on the spindles evenly in a RAID 5 setup, so you do get the increased read speed from all the drives (theres not one "backup drive"). The parity portions are only used during writing or data recovery, so that shouldn't lower read speed. The redundancy will make you lose 1 disc in space though, and writing will almost certainly take longer.

 

[shanghaied]

No no no. You're so half-right that you're wrong.

RAID 5 can allow reads from all spindles simultaneously

So does RAID 0.

3 spindles in RAID 5 will have better performance of two in RAID 0

Not true. At most you get the same performance since only 2 spindles are used to hold data while the third holds a error-correcting check. You generally get slightly worse on reads because of the overhead of the calculations for the third drive. Writes to a RAID 5 are almost always slower since you have to generate the error-correction, so it depends on your controller. Add-in controllers are faster than built-in motherboard controllers.

Depending on the brand and quality of your RAID controller, RAID 5 should be noticably faster than RAID 0

No it won't. If you have RAID 0 all drives are used for data so you get about the number of drives times the speed of a single drive; but for RAID 5 you get the number of drives minus one times the speed of a single drive, so reads are slower for the same number of drives. Also writes are significantly slower, depending on the controller because the controller has to create the extra drive's correction codes.

If gaming speed is your only concern, stick with RAID 0. However, multiple drives mean that one drive failure out of the set will cause the whole array to fail. For example, if you have a 1% chance of failure per drive, four drives in RAID 0 will give you 99%^4 chance of no failure = almost 4% chance of failure instead of 1%. However, with four drives in RAID 5 you need two drives to fail, meaning that your chance of failure means about 1/100 of 1% (= 0.0001) of simultaneous failure times 6 ways to fail means 99.99%^6 or about a .06% chance of loss of the whole array. .06% vs 4% means a factor of 67 difference in reliability in this case for RAID 5 vs RAID 0.

So I'd say, if you have only 2 drives in RAID 0, and you use the RAID only for non-critical data, your drive setup is probably okay. If you're using more drives, consider RAID 5. You won't get the maximum speed, but you won't lose everything to a single crashed hard drive.

--dv

Good for you to correct the previous misconception. I knew it was wrong as soon as I read it. So you saved me the trouble of a response. You know the probability theory better than I do, however. I am planning to install three Hitachi 2 GB 7200 RPM hard drives in Raid 5 on my new system. I am going to use a SSD for the operating system and application programs. Maybe later on, when the price of SSDs come down, I will consider three SSDs in Raid 5. But only if the Trim function works in Raid configuration.

 

[sminlal]

Maybe later on, when the price of SSDs come down, I will consider three SSDs in Raid 5. But only if the Trim function works in Raid configuration.

The jury is, I think, still out on this one yet - but the general consensus seem to be that redundancy protection for SSDs is much less important than for hard disks. SSDs are not prone to data loss - their most common failure mode is the inability to write new data. Thus, the theory goes, it should be easy to recover the data off a "failed" SSD.

That having been said, for a high-uptime configuration failure to write data may be just as critical. And, I have to imagine that a failure to write data at the wrong time and place (during a file system metadata update) could render at least some data unreadable, or at least a lot harder to access without recovery tools.

 

[shanghaied]

Now that I think about it, I believe I recall SSDs have some sort of built-in error correction already, so it would not make sense to put SSDs into a RAID 5 configuration.

 

[shanghaied]

Actually, after reading additional material, I think RAID 10 is the way to go for both performance and redundacy, but a little expensive. Disk drives are pretty cheap these days. But most likely I will just install RAID 0 for my hard drives and perform backups on left-over external disk drives just like I do now.

 

[cjl]

Wow - thread necromancy alert...

 

[AshcanPete]

Good for you to correct the previous misconception. I knew it was wrong as soon as I read it. So you saved me the trouble of a response. You know the probability theory better than I do, however. I am planning to install three Hitachi 2 GB 7200 RPM hard drives in Raid 5 on my new system. I am going to use a SSD for the operating system and application programs. Maybe later on, when the price of SSDs come down, I will consider three SSDs in Raid 5. But only if the Trim function works in Raid configuration.

Good for you to correct the previous misconception. I knew it was wrong as soon as I read it. So you saved me the trouble of a response. You know the probability theory better than I do, however. I am planning to install three Hitachi 2 GB 7200 RPM hard drives in Raid 5 on my new system. I am going to use a SSD for the operating system and application programs. Maybe later on, when the price of SSDs come down, I will consider three SSDs in Raid 5. But only if the Trim function works in Raid configuration.

Yes, you are definately right about Raid 0 being faster (especially with write speeds), and also allows for more storage space. In other words:

Raid 0 speed = (speed of one drive) x (number of drives in array) [usually not quite this fast]
Raid 5 speed = a bit less than raid 0 on read (due to complexity); much slower on write (due to parity)

Raid 0 failure rate = (failure rate one drive) x (number of drives)
Raid 5 failure rate = approximately zero

In fact svyerkgeniiy was a little off in probablity of failure rates above, raid 5 is even more reliable than he states. In his example, if you were to have a 1% chance of failure per drive (assuming he means over an average lifespan of around 2 years), then for normal drives you have a 1% chance data loss. For raid 0 with 3 drives you have a 3% chance.

With raid 5, you have (3% chance for one drive to fail) x (chance that a second drive fails BEFORE YOU REPLACE THE FIRST DRIVE) which is a very low chance. So if you assume it takes you a day to replace the drive (or backup data), the raid 5 failure rate would be 3% x [2%/(365x2)] = 0.00008% chance of data loss. So raid 5 would actually be 36500 times more reliable than raid 0.

So my point is that raid 5 (or pretty much any redundant raid array) not only makes data loss unlikely, it makes it virtually impossible (this is only for data loss from hard disk failure however).

Overall, raid 0 with 3 drives is super fast yet somewhat risky, raid 5 with 3 disks is kinda fast but very reliable. I actually prefer raid 0, its worth the speed! (just backup your important files elsewhere)

 

[cipher_nemo]

I'm necroing this thread since it's a high-ranking Google result and might mislead others researching RAID 5 vs. RAID 0 speeds for 3 or more drives.

Raid 0 speed = (speed of one drive) x (number of drives in array) [usually not quite this fast]
Raid 5 speed = a bit less than raid 0 on read (due to complexity); much slower on write (due to parity)

This is close, but not completely true. Writing data on a RAID 5 is NOT "much slower" than RAID 0 if you take the RAID controller out of the equation. An integrated RAID controller (ie: Fake RAID or "HostRAID") that relies about the system's CPU for computations will have much slower writing than RAID 0. But a hardware-based RAID, such as a $200+ Areca, HighPoint, Adaptec, etc. card (look for cards with heatsinks on the RAID chip) will have have RAID 5 writes that approach RAID 0 write speeds. RAID 5 always takes more calculations, but for hardware-base RAID cards, this can be very close to RAID 0 speeds.

Most $200-something hardware-based RAID cards will take a minor write performance hit in RAID 5 when compared to RAID 0. Most $350+ hardware-based RAID cards with on-board cache (ie: 256MB, 512MB, 1GB, 2GB, etc.) will have near-RAID 0 write performance, if not identical performance. All of that is taking into account the subtracted parity drive which you sacrifice in the name of reliability if 1 drive fails.

Bottom line:

For built-in RAID controllers (ie: "fake RAID", ICH9, ICH10, etc., etc.), always run RAID 0 for performance, RAID 1 for redundancy, or RAID 01 or 10 for a mixture of both. Don't do RAID 5 (or even 3, 4, or 6) on built-in RAID controllers due to the severe performance hit.

For hardware-base RAID controllers try to do RAID 5 for 3 drives, RAID 5, 01, or 10 for 4+ drives. RAID 6 is fine for extra redundancy on uneven drive numbers of 6+, but I'd rather do RAID 5 for performance or RAID 01 or 10 for redundancy if you're doing 6+ drives.

 

[Nudebie]

So I'm just setting up 4x 1.5tb Seagate 7200rpm drives in a Raid5(ICH10) array. From all the postings I have not found the answer to this: Will the Raid5 read and write speeds be slower then just using one drive in non-raid mode???

 

[cipher_nemo]

So I'm just setting up 4x 1.5tb Seagate 7200rpm drives in a Raid5(ICH10) array. From all the postings I have not found the answer to this: Will the Raid5 read and write speeds be slower then just using one drive in non-raid mode???

RAID 5 on three drives with a built-in RAID controller on a motherboard will have significantly slower write speeds than just a single drive. Read speeds will most likely be significantly higher, to the tune of may be 1.5 times the speed of a single drive.

In all honestly, don't do RAID 5 on a built-in RAID controller (ICH10). If you have three 1.5 TB drives and want redundancy as well as performance, either buy one more and do a RAID 10 or choose RAID 0 or 1.

If you get a fourth drive, I'd personally put two into a RAID 0 for the operating system and applications, and a RAID 1 of the other two drives for your data and files where you want some redundancy. And if you want to squeeze a little more performance out of your RAID 0 of the two drives for your operating system, "short-stroke" them (make each 1/10 to 1/3 of the total size only). If you short-stroke two 1.5TB drives in RAID 0 (which would total 3GB) you can still get plenty of room for an O/S and applications with faster seek times.

http://www.tomshardware.com/reviews/short-stroking-hdd,2157.html

You don't need tools or firmware changes to short-stroke. You can always opt to use less space on each drive when setting up a RAID array or make a small partition since the faster parts of the drive (inside of the platters) is always used first.

 

[Nudebie]

Okay, it's back to the drawing board then. This is now an official call for advice since this is my first raid attempt.

Objective: Primarily a DVR machine using AVerMedia's NV8416E4 camera capture cards recording 16-32 x 30 fps at 720X576 pixels plus playback without interupting recording. Plus one or two 360 megapixel ip cams. http://www.avermedia.com/AVerDiGi/Product/Detail.aspx?id=17516

Machine: Asus P6X58D-E, i7 960, Corsair DDR3-1333 12GB, Radeon HD 5770, 5x Barracuda 7200 32mb 1.5TB drives, and a LG Bluray writer.

Current configuration: One hd with a 350gb partition for os and the rest of that drive for general storage and the LG Bluray writer both plugged into the on board Marvell Sata 6Gb/s controller. The other 4 hds plugged into the on board ICH10R Southbridge ports and set in bios as raid. I need these exclusivly for the video record/playback.

I originally set the raid to 0 and waited all friggin night for Win7 Ultimate to create a simple disk. Then I started to worry about disk failure and down time. Can't afford the cameras down for long. I figured if I went raid5 it could still hobble along while rebuilding the new drive.

The video gets overwritten as the drives fill so losing the video is not as bad as a lengthy down time to rebuild. It sorta would be good to recover without down time. So what is the best raid for my purpose?

 

[Nudebie]

Okay, decided to put the 4 drives in raid10 for video storage. Thanks.