RAID 5 vs 10

  • Thread starter Deleted member 88227
  • Start date
D

Deleted member 88227

Guest
Which one would be the best suited for gaming?

I'm not 100% sure on how RAID 5 works, but I know it requires at least three disks to operate, should one fail no data is lost, you'll just need to replace the drive and let it rebuild. What about the performance side? Read/Write.

I know how RAID 10 works, it's essentially RAID 1 and RAID 0 combined? Could this technically be done with just three drives or does their have to be an even number of drives? Lets say you have two 1GB drives and one 2GB drive. Can RAID 0 = two 1GB drives and RAID 1 = (two 1GB drives) + (2GB drive)? If that's the case, how can that be faster? I mean, the data still has to be written to the single drive, so does that mean only reading data is faster? Would this not make RAID 5 better. but how?

It's so confusing. :confused:
 
Why don't you buy 1 SSD for your OS/games and then 2 large 7200 RPM drives, one for storage and the other for backup (assuming your talking about RAIDing spinning disks and not SSDs).

You need at least 4 drives for RAID10. RAID 5 is faster for large sequential reads/writes (video streaming). RAID 10 is faster for small random reads/writes...........which is typically the bottleneck for everyday systems.

RAID 5 rebuild time is horrible compared to RAID10, should you lose another disk during that rebuild time, your screwed............well I guess not if you're doing your backups like you should be.
 
Why don't you buy 1 SSD for your OS/games and then 2 large 7200 RPM drives, one for storage and the other for backup (assuming your talking about RAIDing spinning disks and not SSDs).

You need at least 4 drives for RAID10. RAID 5 is faster for large sequential reads/writes (video streaming). RAID 10 is faster for small random reads/writes...........which is typically the bottleneck for everyday systems.

RAID 5 rebuild time is horrible compared to RAID10, should you lose another disk during that rebuild time, your screwed............well I guess not if you're doing your backups like you should be.

There are some in accuracies here so let me clarify:

Assuming you are not using software raid and a good controller then raid10 will be faster for reads and slower for writes than raid5 (sequential reads/writes). Some controllers/software raid10 will not read from all the disks when doing reads (only half) so that is something to watch out for as it means your read performance will be half when in reality on a good controller raid10 read performance should be as good as raid0.

Raid10 write performance will be exactly what raid0 would be with half the number of disks so assuming a good controller raid5 will give you superior sequential write speed on an array with the same number of disks.

Because data is also split between all drives in the member set raid5 read performance for random/small writes should be the same that raid10 is. Sequential reads on raid5 will be slower (assuming the same number of disks and no controller bottleneck) due to the drives needing to skip over parity stripes when doing reads. Small random writes is where raid5 really hurts. Generally you will not do a lot of these and if you use a controller with a BBU and a good amount of cache then you should not notice low performance on small random writes during normal operations (if your running a mysql server where its doing small random writes all the time this would be another story).

When raid5 does writes below the stripe size data must first be read off the drives in order to calculate the parity of the partially written stripes which cuts the iops in half as it requires two operations instead of just strait writing.

Another downside to raid5 is when the array is degraded. Because data must be re-constructed from parity (and to do so you have to read from all disks) means your iops goes down to one disks worth when your array is degraded so degraded raid5 performance is significantly worse than raid10 if your doing a lot of read operations. Simply sequential read/writes wont actually suffer that badly if you have a lot of drives.

As for raid 5 taking a lot longer to rebuild that is completely untrue. Again assuming a good controller you will be bottlenecked by the destination disks write speed. In fact its more likely that raid5 rebuild speed wont slow down as much as raid10 due to the fact that the data is being reconstructed from doing reads across all the disk in the array where as in typical raid10 configurations all the reads are coming from the single other disk in the mirror and it can slow down a lot just with a little disk activity.

With 8 disks or more I suggest going raid6 for the added protection. I would say from a data security point that raid10 is indeed less likely to have the array lost (or corruption) during a disk failure than raid5 is but raid6 is the most secure when it comes to data/volume loss or file-system corruption during a rebuild and that is what I try to use on all my systems now.

If you mainly want the raid for performance (and not disk space) I think I would go raid10 if using 6 drives or less but if your main reason for the raid is increased capacity (and speed is also an important factor) then I would go raid5 (or raid6 if using more than 6 disks).
 
There are some in accuracies here so let me clarify:

Assuming you are not using software raid and a good controller then raid10 will be faster for reads and slower for writes than raid5 (sequential reads/writes). Some controllers/software raid10 will not read from all the disks when doing reads (only half) so that is something to watch out for as it means your read performance will be half when in reality on a good controller raid10 read performance should be as good as raid0.

Raid10 write performance will be exactly what raid0 would be with half the number of disks so assuming a good controller raid5 will give you superior sequential write speed on an array with the same number of disks.

Because data is also split between all drives in the member set raid5 read performance for random/small writes should be the same that raid10 is. Sequential reads on raid5 will be slower (assuming the same number of disks and no controller bottleneck) due to the drives needing to skip over parity stripes when doing reads. Small random writes is where raid5 really hurts. Generally you will not do a lot of these and if you use a controller with a BBU and a good amount of cache then you should not notice low performance on small random writes during normal operations (if your running a mysql server where its doing small random writes all the time this would be another story).

When raid5 does writes below the stripe size data must first be read off the drives in order to calculate the parity of the partially written stripes which cuts the iops in half as it requires two operations instead of just strait writing.

Another downside to raid5 is when the array is degraded. Because data must be re-constructed from parity (and to do so you have to read from all disks) means your iops goes down to one disks worth when your array is degraded so degraded raid5 performance is significantly worse than raid10 if your doing a lot of read operations. Simply sequential read/writes wont actually suffer that badly if you have a lot of drives.

As for raid 5 taking a lot longer to rebuild that is completely untrue. Again assuming a good controller you will be bottlenecked by the destination disks write speed. In fact its more likely that raid5 rebuild speed wont slow down as much as raid10 due to the fact that the data is being reconstructed from doing reads across all the disk in the array where as in typical raid10 configurations all the reads are coming from the single other disk in the mirror and it can slow down a lot just with a little disk activity.

With 8 disks or more I suggest going raid6 for the added protection. I would say from a data security point that raid10 is indeed less likely to have the array lost (or corruption) during a disk failure than raid5 is but raid6 is the most secure when it comes to data/volume loss or file-system corruption during a rebuild and that is what I try to use on all my systems now.

If you mainly want the raid for performance (and not disk space) I think I would go raid10 if using 6 drives or less but if your main reason for the raid is increased capacity (and speed is also an important factor) then I would go raid5 (or raid6 if using more than 6 disks).

Innacuracies?! I would like to see a single credible source stating RAID 5 rebuild time is better than RAID 10. I'm not talking about using pro-active hot spares either. I'm talking you yank one drive out of the array and force it to rebuild.

You ought to get up to speed on Software RAID, this is the new millenium where we have fast CPU's. To expand on this comment: yeah a dedicated RAID controller is going to perform better, but is it worth the cost especially for a gaming rig? If we're talkig MDADM, the gap narrows.

8 disk RAID6 for a gaming machine.....that's real good advice there.
 
Last edited:
8 disk RAID6 for a gaming machine.....that's real good advice there.

That's not what he said. With RAID-10, you lose the wrong two disks, and you lose your array. With RAID-6, you can lose any two disks and still have your array survive. The more HDs you have, the more statistically likely you are to have two disks fail, making RAID-6 a better alternative.

But why is an eight-disk RAID-6 bad advice for a game machine? I have 10 drives in my gaming rig, with room for four more. Assuming I ever fully populate it with 14 drives, what's you RAID recommendation for me?
 
Innacuracies?! I would like to see a single credible source stating RAID 5 rebuild time is better than RAID 10. I'm not talking about using pro-active hot spares either. I'm talking you yank one drive out of the array and force it to rebuild.

You ought to get up to speed on Software RAID, this is the new millenium where we have fast CPU's.

8 disk RAID6 for a gaming machine.....that's real good advice there.

This is why I asked if capacity was a reason on whether or not for the raid. If he really doesn't need a lot of disk space then likely not using raid at all and just running off an SSD would be a better option. Guess what, I have 20 disks in raid6 and my machine is also a gaming machine but I also happen to have a massive need for storage as well.

Most of my knowledge when it comes to raid performance is not from what I have read but from what I have actually experienced. I have seen very many controllers that will rebuild raid5 or raid6 at the *speed* of the drive. For example on my sisters computer that had 3x 500GB in raid5 on even a cheap perc5 it only took about 2.5 hours to rebuild which is pretty much how long it takes just to write zero's to the drive at the drives max write speed.

I can tell you write now that rebuild speed of raid6 on an ARC-1222 with 8x1 TB drives and raid10 (with same drives) was just about identical.

Also software raid still isn't as fast hardware raid on a good controller. The only good implementation of software raid I have seen is on ZFS/solaris and even then it does not perform as good as a hardware controller and it was just about maxing out CPU usage coming close (writes at about 60-70% the speed of a hardware raid controller).
 
That's not what he said. With RAID-10, you lose the wrong two disks, and you lose your array. With RAID-6, you can lose any two disks and still have your array survive. The more HDs you have, the more statistically likely you are to have two disks fail, making RAID-6 a better alternative.

But why is an eight-disk RAID-6 bad advice for a game machine? I have 10 drives in my gaming rig, with room for four more. Assuming I ever fully populate it with 14 drives, what's you RAID recommendation for me?

Show me how much you spent on your drives and RAID controller and let's compare that to a good SandForce SSD supplemented by a decent performing 7200 rpm drive for storage of "I don't need to open this *.doc, *.jpg file with blazing fast speeds".
 
So will raid5... any raid rebuild will slow down with disk activity.

Think of it this way... Your array is under enough load that all the drives in the array are being bogged down about 50% and thus can't read more than 50% of their normal read speed.

On raid10 this of course means that the only drive you are able to read off of for the rebuild is only providing you with 50% of the read speed and thus will slow down your rebuild due to a read bottleneck (most mechanical disks these days can write at about 90% of their read speed).

Now consider the same on raid5. Even if your entire array is bogged down by 50% it is reading from all the disks to reconstruct the data. Say you have 6 disks in raid5, well instead of 100% reading from one drive you are now doing about 20% reads from the other 5 disks to reconstruct the failed drive, well even if each drive is bogged down about 50% in the array it can still provide 20% of additional reads which still gives the destination (rebuilding drive) 100% of its write speed.

In the real-world there will likely not be a huge difference I am just saying that raid5/raid6 has the potential to be less affected by disk activity during the rebuild. It really depends a lot on how harsh the disk activity is as well whether or not the raid controller is not gonna care and slow things down enough and rebuild the drive as fast as it can. If you think about it logically you can see how (again assuming your not bottlenecked by the card) that rebuilding raid5 can definitely be faster than raid10.

Also I will just say it now that I have only ever seen one case in which raid10 has out-survived raid6 (a 8 disk machine with 3 failed drives and still working) but I have seen about 8-9 raid10 arrays fail where they would have been ok if they were raid6.
 
Also just to add the biggest reason for going raid10 is simply that you don't need to buy a 'good' raid controller like you pretty much need to if you dont want shit performance with raid5. I am willing to bet a lot of the cheap controllers that will do raid10 also will not read from all 4 drives when doing sequential reads though.
 
Also just to add the biggest reason for going raid10 is simply that you don't need to buy a 'good' raid controller like you pretty much need to if you dont want shit performance with raid5. I am willing to bet a lot of the cheap controllers that will do raid10 also will not read from all 4 drives when doing sequential reads though.

Price of a controller unfortunately isn't an indicator of its parity RAID performance. 3ware 8506-12 cards have been ~$30 recently on Ebay, and they beat anything I've ever paid three digits for. (I'll never pay three digits for a card again)

3ware with its TwinStor tech is the only card I'm aware of that will use all drives when doing reads in RAID-10. Apparently some form of software RAID in some Linux flavor's also capable of this.
 
Price of a controller unfortunately isn't an indicator of its parity RAID performance. 3ware 8506-12 cards have been ~$30 recently on Ebay, and they beat anything I've ever paid three digits for. (I'll never pay three digits for a card again)

3ware with its TwinStor tech is the only card I'm aware of that will use all drives when doing reads in RAID-10. Apparently some form of software RAID in some Linux flavor's also capable of this.

Actually I am pretty sure TwinStor does not do anything for sequential reads on raid10. I can tell you from experience that 3ware 9650se cards will not read from all disks in raid10 when doing sequential reads. Basically now and then they alternate which disk they are reading from in the pair every couple of seconds or so but they never read from all the disks.

LSI is the same way as well as linux mdadm. The only controller I know that will show solid disk activity LED's for all disks in the raid set (and gave the fastest read benchmarks) are areca controllers. I only have first and raid10 experience with 3ware, LSI, areca, and linux mdadm so I can't say the same is true for all the other ones.

I am pretty sure that when you have a lot of outstanding I/O the controllers will read from all the disks so doing multi-threaded or asynchronous I/O with lots of random reads you will be reading from all the drives but they wont do it on a single threaded sequential read.

EDIT:

Also comparing old PCI-X controllers to new PCI-E controllers is not exactly comparing apples to apples either. Performance will be horribly bottlenecked if they are used on regular PCI so that controller is not an option for most people as few people have motherboards with PCI-X support.
 
Last edited:
Actually I am pretty sure TwinStor does not do anything for sequential reads on raid10. I can tell you from experience that 3ware 9650se cards will not read from all disks in raid10 when doing sequential reads. Basically now and then they alternate which disk they are reading from in the pair every couple of seconds or so but they never read from all the disks.

Supposedly it does. They read from whatever disk will give them the best access at that moment. I know from experience this works well for random reads, but yeah I'll give the "supposed" remark to this, even though their literature says it does.

I am pretty sure that when you have a lot of outstanding I/O the controllers will read from all the disks so doing multi-threaded or asynchronous I/O with lots of random reads you will be reading from all the drives but they wont do it on a single threaded sequential read.

If you want to read two files seq with TwinStor, it has the potential to see your RAID-10 as two individual RAID-0s and read the two files at full seq speed. If other controllers are doing this, I wasn't aware of it. Though all controllers should be capable of this anyway, considering how much they cost.

It's not like RAID is rocket science, although 3ware at times a decade ago would make you wonder with some of their firmware bugs.
 
I'm relatively sure that a recent mdadm will read from all disks in a RAID10 setup during one sequential read. In RAID1 mode a single read will only go to one disk, while multiple sequential reads will get balanced between disks.
 
My RAID 10 vs RAID 5 results on a Dell R710 with both arrays degraded. This is on an H700 controller with both file systems partition-aligned to 1 meg and formatted with 64K File Allocation Unit Size and using a 64K stripe. The RAID 10 degraded actually performs better than a healthy RAID 10 since essentially it's now RAID 0. This isn't even taking into account the tremendous hit that would occur if each were rebuilding.

Raid 10 vs RAID 5 degraded:

http://filebin.ca/rjzmzg/RAID10vRAID5_degraded.htm

Below is RAID 10 vs RAID 10 degraded:

http://filebin.ca/fxfxdx/RAID10vRAID10_1DEGRADED.htm
 
Last edited:
EDIT:

Also comparing old PCI-X controllers to new PCI-E controllers is not exactly comparing apples to apples either. Performance will be horribly bottlenecked if they are used on regular PCI so that controller is not an option for most people as few people have motherboards with PCI-X support.

If you're doing two reads on both types of controllers, and the PCI-X one is on a PCI slot but sees the RAID-10 as two RAID-0 arrays while the PCI-E sees it as one RAID-10 array so must go to non-seq reads, the PCI-E card will be the bottleneck.

But I've never found PCI to be a horrible bottleneck. RAID's mainly done for the redundancy, and having the PCI-X card in a PCI slot doesn't hinder the card in init/repair/rebuild runs.
 
Show me how much you spent on your drives and RAID controller and let's compare that to a good SandForce SSD supplemented by a decent performing 7200 rpm drive for storage of "I don't need to open this *.doc, *.jpg file with blazing fast speeds".

That's not a RAID recommendation at all. I want some redundancy while having storage.

But to answer your question: I spent nothing for the RAID controller, and the 2TB HDs cost $60. You won't beat that price for storage space.
 
That's not a RAID recommendation at all. I want some redundancy while having storage.

But to answer your question: I spent nothing for the RAID controller, and the 2TB HDs cost $60. You won't beat that price for storage space.

So this is all relating back to the OP, asking for opinions on RAID for a gaming rig. Maybe I should have ignored the whole reason for RAID as I don't see most gaming rigs as critical uptime devices............."I really need to be able to play BFBC2 with zero downtime!", which is why alot of gamers just go RAID0 with healthy backups.

As for prices, ebay and "my bro handed it down to me" aren't really valid. We're assuming the OP has to buy this stuff brand new and has to get it via retail.

So my RAID recommendation for the OP is none unless he's going to reply back with why he needs RAID for something else other than gaming, like a media server that can't have any downtime.
 
Wow lots of good reading. I hate down time, just as much as the next person, but it's not that big of a deal if I have to wait hours for a RAID array to rebuild as I work at least 8 hours 5 days a week. With that being said, I want to see the performance in gaming applications mostly. I also use my computer for photoshop work and in the future I plan to be doing some HD video editing, but I think a RAID setup that benefits gaming will also benefit my occasional photography/videography work. I would get an SSD, but I would much rather spend the money on more space.

Does RAID 5 require 6 drives or can it be done with less?

I really do not know which RAID setup I should go with, I was just gonna go with RAID 0 and keep up with my backups, but I figured why not go with something that can do both.

The controller I plan to use is built into the motherboard I will be purchasing:
ASUS P7P55D-E LGA 1156 Intel P55
 
I really do not know which RAID setup I should go with, I was just gonna go with RAID 0 and keep up with my backups, but I figured why not go with something that can do both.

RAID is not a backup. regardless of what RAID level you pick, you will always need to do backups. RAID does not protect against flood, fire, theft, lightening strikes, viruses, data corruption (for the most part), solar flares, acts of various gods, hackers, the CIA, etc.

all RAID does do is ensure that if a single hard drive kicks off, you wont be completely dead in the water, and that the computer will still function to a degree while you fix or replace the dead disk.
 
I fully agree with what ghost6303 said in the last post. If you have data that you can not or do not want to risk loosing make backups of it. RAID does not remove this requirement.
 
Last edited:
That's not what he said. With RAID-10, you lose the wrong two disks, and you lose your array. With RAID-6, you can lose any two disks and still have your array survive. The more HDs you have, the more statistically likely you are to have two disks fail, making RAID-6 a better alternative.

But why is an eight-disk RAID-6 bad advice for a game machine? I have 10 drives in my gaming rig, with room for four more. Assuming I ever fully populate it with 14 drives, what's you RAID recommendation for me?

Is that RAID array for your OS or for storage?
Why not just get a SSD for the OS, then use a RAID array for storage?
 
I fully agree with what ghost6303 said in the last post. If you have data that you can not or do not want to risk loosing make backups of it. RAID does not remove this requirement.

Exactly, RAID may be robust, but it's not a fail-safe. Backing up with optical media and HDDs is great for home/SOHO use, and for mission critical and enterprise backup, use tape drives/libraries.
 
Back
Top