What Is the Advantage of Raid 5 Over Raid 1
- Home
- Storage & Backup
- Data Storage
I'm about to buy some new servers. These will be used for a fileserver, webserver, SQL, and Exchange.
Looking at RAID levels, it seems that RAID 1 or 0+1 might be best given the low cost of drives.
I think RAID 5 was meant to maximize space and minimize cost, but with a performance sacrifice.
Am I just misinterpreting what I read on them?
RAID1 - Mirroring will be slightly more performant then a single drive as it can read off of both drives, but the writes are longer.
RAID5 outperforms RAID1 in both reads and writes. It does maximize space as you only loose 1 drive to parity, no matter the size of the array. This is by far the most common RAID level.
RAID1 + 0 (commonly called RAID ten) out performs all of them by a good margin. Since it's RAID1 with stripping it is also the robust and can suffer a higher level of failure. Mostly. You need a minimum of 4 disks and you'll loose 2 of them to redundancy. Why I say mostly is, let's say you have a 8 disk RAID10 set. You could theoretically loose 4 disks all at once and still have the array up and working, as long as none of the failed drives are in the same RAID1 set.
If you've got the money to get the biggest drives and you want performance, RAID10 is the best way to go. RAID5 is next down, especially if your RAID controller will support a hot spare.
Stay away from RAID6 as it really hits you on the write performance for a dubious extra redundancy.
RAID0 should be avoided at all costs.
- ROM
- CPU
- RAM
- GPU
21 Replies
RAID1 - Mirroring will be slightly more performant then a single drive as it can read off of both drives, but the writes are longer.
RAID5 outperforms RAID1 in both reads and writes. It does maximize space as you only loose 1 drive to parity, no matter the size of the array. This is by far the most common RAID level.
RAID1 + 0 (commonly called RAID ten) out performs all of them by a good margin. Since it's RAID1 with stripping it is also the robust and can suffer a higher level of failure. Mostly. You need a minimum of 4 disks and you'll loose 2 of them to redundancy. Why I say mostly is, let's say you have a 8 disk RAID10 set. You could theoretically loose 4 disks all at once and still have the array up and working, as long as none of the failed drives are in the same RAID1 set.
If you've got the money to get the biggest drives and you want performance, RAID10 is the best way to go. RAID5 is next down, especially if your RAID controller will support a hot spare.
Stay away from RAID6 as it really hits you on the write performance for a dubious extra redundancy.
RAID0 should be avoided at all costs.
What you want. is Raid 1+0 ie Raid 10. Mirrored THEN striped. So if you lose a drive you don't lose your stripe set. if you're in need of only OS drives. Then getting two drives and mirroring them is faster and a better solution than getting 3 drives and doing a raid 5.
If you are looking to do local storage and buy a whack of drives though. Raid 10 is MUCh more costly than Raid 5 is. But keep in mind, a dual disk failure is scarily quite common with Raid 5. at which point you lose everything. This is why people do Raid 6. Having two parity drives.
Ahh you mean RAID 1+0, or RAID 10. That is probably the best bet if you don't want to go with RAID 5.With a 4 drive RAID 1+0, you'll get read performance of 2x-4x a single drive, while writes will be from 1x-2x. In theory, that is. In reality, if using a RAID PCI card or motherboard solution hooked to the south bridge, you'll most likely max out the read speed.
Do some googling around and you ought to find whats right for you.
Thanks for the response. My understanding of RAID appears to have been incorrect.
A little further googling did turn up this MSDN article.
http://msdn.microsoft.com/en-us/library/ms190764.aspx?ppud=4
It seems to echo that RAID 5's best as far as unwasted space and speed, and that 1+0 performs better. I'm going to read on and try to wrap my head around why RAID 1 wouldn't perform as well as 1+0.
I'm also a RAID 10 fan, with drives being such low cost items now. Raid 10 outperforms RAID 1 because it also incorporates striping. RAID 5 stripes the data as well, but there is a performance hit (especially on writes) because of the parity calculations.
Also, the RAID controller chipset can make a big difference in performance and reliability. Something with the Intel IOP348 chip would be my recommendation.
You can read my input on RAID 5 in "The Agony and Ecstasy of RAID" on Datamation. I am also a signatory of BAARF (The Battle Against All Raid Five).
Raid 1 isnt' as fast as Raid 10 because of Spindles. It's like saying why isn't one guy as fast as two guys working... Raid 10 assumes you have 2 guys (mirrored for redundancy) who are working on something at the same time splitting the work.
Stripping means the data is sent not just to one drive to write, but to the other drive also. So both drives are writing a portion at their top speed. Rather than a single drive writing the whole thing at its best speed.
David1618 wrote:
RAID 5 is best in unwasted space while still having some about of redundancy. It is NOT best in speed. BAARF was actually formed to fight the misconception that RAID 5 had any performance advantage in database systems because that used to be a common notion but studies have shown that the parity calculation just kills the performance too much for it to ever be chosen for that purpose.Thanks for the response. My understanding of RAID appears to have been incorrect.
A little further googling did turn up this MSDN article.
http://msdn.microsoft.com/en-us/library/ms190764.aspx?ppud=4
It seems to echo that RAID 5's best as far as unwasted space and speed, and that 1+0 performs better. I'm going to read on and try to wrap my head around why RAID 1 wouldn't perform as well as 1+0.
Thanks for the article link. I still don't totally get the whole thing.
Mainly I don't get this sort of situation... Picture a system of 4 drives named A B C and D, in various configs, except in the RAID 1 example where there's only drives A and B.
I've got 2 blocks of data to write.
In RAID 0, that is written to block A1 and B1, because it is striped. This should be fast because the write operation should be able to go to both drives at once.
In RAID 1 that goes out to 2 drives, and gets written to blocks A1 and A2, mirrored to B1 and B2. 2 writes to 2 drives should take twice as long as 1 write to 2 drives.
In RAID 10 that write operation would go out to 4 blocks on 4 drives, A1 and C1, being mirrored to B1 and D1. I can see where writing one block to 4 drives might be faster than writing 2 blocks to 2 drives, but don't know how it is faster than RAID 0.
In RAID 5 it sounds like that 2 blocks would get written to blocks A1 and B1, and a parity bit would be calculated, found on drive D, then overwritten. This just sounds slow.
Oh well, I've not been happy with RAID 5's performance on what should be a speedy system, so I'm just gonna give 1+0 a shot.
On another note, your article, Scott, indicated that Mirroring's not a backup. Typically it isn't, but I used to have a few spare hotswap drives in my last shop. I made my OS drives mirrored drives, and would rotate those monthly. Once that really saved my butt. I screwed up a system, which got mirored of course. I just shut down, put in the one good month-end drive, and booted up. I had to make a new account or 2 that wasn't on the old drive, but I was in good shapee otherwise.
Again thoughg, thanks for all the info, have a good weekend all!
The problem with RAID0 is not speed, it's redundancy. Since data is stripped across the 2 drives, you're actually 50% less reliable then a single drive. In RAID0 if you loose 1 drive in the set, the entire set is lost.
Interestingly, Maximum PC did some gaming tests of a single drive versus a RAID0 and found the single drive performed at the same level. This was attributed to the overhead of running the RAID. So, in my book, I'd rather run a single drive then a RAID0 set.
RAID10 benefits from having so many heads reading and writing at the same time. The write head on the drive that's going to receive the next block to write can actually pre-position itself to be ready for the write operation. Read has the same advantage and since you have more heads doing the reading (4 instead of 2) your performance goes up.
RAID 0 is definitely the fastest, you are correct. It is not always considered to be true RAID, though, as it is not a redundant array but simple an array of disks (AID).
David1618 wrote:
In RAID 1 that goes out to 2 drives, and gets written to blocks A1 and A2, mirrored to B1 and B2. 2 writes to 2 drives should take twice as long as 1 write to 2 drives.
RAID 1 does not work the way that you describe. Drive writes are not done sequentially but concurrently. If both drives are identical and there is no error in the write process then the write time of two drives is identical to the write time of a single drive. However the read time is done just like RAID 0, in parallel, so you get roughly double the read performance with RAID 1 than you do with a single drive.
David1618 wrote:
That's an interesting use of mirroring. Yes, technically you were mirroring the drives. But also once you pull the drive out it is no longer mirrored (since any further writes don't go to all drives) so at that point you could consider it a backup since the mirroring is no longer active. It is the active act of mirroring that makes it not backup (as it causes them to fail together.) In the way that you are using it, it is a one time copy rather than a mirrored drive - although if you left it in it would be a mirrored drive. So it is a bit of a hybrid model.On another note, your article, Scott, indicated that Mirroring's not a backup. Typically it isn't, but I used to have a few spare hotswap drives in my last shop. I made my OS drives mirrored drives, and would rotate those monthly. Once that really saved my butt. I screwed up a system, which got mirored of course. I just shut down, put in the one good month-end drive, and booted up. I had to make a new account or 2 that wasn't on the old drive, but I was in good shapee otherwise.
And a pretty hefty performance impact as the new drive would have to be completely brought up to parity with the original drive. I wonder what the manufacturer would have to say about using the hot swap slot so frequently too?
True, that would take quite the toll on the hot swapped port. They really are not designed to do that at all. I've definitely seen backplanes fail from overuse like that.
Another consideration from the question RAID 1 vs RAID 5 must also be the specific hardware RAID controllers. Are the RAID controllers supplied with memory and battery backup as well as the types of RAID that they support.
Generally now when looking at any server that is bought I look to try and ensure that the controller card has a minimum of RAID 0, 0+1, 1+0, RAID 5, RAID 6 (Don't tend to use this), JBOD (Not that this really matters).
As usual, Scott is dead on. I'll add another plus for RAID 1. RAID sets can become damaged, and with RAID 1 I have the added luxury of connecting either of the drives to another system and reading the data. This has saved our collective bacon more than once.
Death to RAID 5!
Valley IT Support, Inc is an IT service provider.
My 2 cents, I deploy all my servers unless other wise asked in this config.
4 Drives
First 2 drives in RAID1 as OS part
Second 2 drives in RAID1 as data part.
Why #1, becuase as Mike at SCA stated you can connect to another system and read the drives or I have found connect to the SATA ports on the motherboard and boot the server and all is well!
Why #2, I had a client with a RAID 5 setup and the controler failed and guess what the new controler could not read the existing RAID setup becuase it was on the failed controler. I understand new controlers save the a copy of the conf on the drives but I have yet to test this. And in any case my clients typical do not want to wait the 24hrs for a replacement controler. So this config is also part of my DR planning.
Owenmpk wrote:
RAID 1 is like a fuzzy fleece blanket that makes you sleep well at night. You never hear horror stories about RAID 1, never. Not the most performant but dual RAID 1 does have queue/contention advantages over RAID 10 that does give it slightly better than expected performance, especially if there is a lot of dualing sequential writing going on.My 2 cents, I deploy all my servers unless other wise asked in this config.
4 Drives
First 2 drives in RAID1 as OS part
Second 2 drives in RAID1 as data part.
Why #1, becuase as Mike at SCA stated you can connect to another system and read the drives or I have found connect to the SATA ports on the motherboard and boot the server and all is well!
Why #2, I had a client with a RAID 5 setup and the controler failed and guess what the new controler could not read the existing RAID setup becuase it was on the failed controler. I understand new controlers save the a copy of the conf on the drives but I have yet to test this. And in any case my clients typical do not want to wait the 24hrs for a replacement controler. So this config is also part of my DR planning.
RAID 1 @ 500gb, the actual file server. Mirrored to a 4 x 500gb RAID 10 server with incremental backups. Each in a different office on the same LAN. Use a USB HDD to take copies offsite.
Just lost a RAID 5 array, cost $13,403 to get the data back. I will only use RAID 1 and 10 from now on. Period. And Hitachi drives.
This topic has been locked by an administrator and is no longer open for commenting.
To continue this discussion, please ask a new question.
What Is the Advantage of Raid 5 Over Raid 1
Source: https://community.spiceworks.com/topic/81351-raid-1-vs-raid-5