ANOTHER RAID 5 QUESTION

[bluefmc]

my employer is ordering a dell server running redhat linux with hardware based scsi raid 5. i have a few questions concerning raid 5 arrays. i know 3 disks is the minimum number of disks for a raid 5 array, but i was wondering if a 3 disk array sees any performance benefits over just a single disk. what would be the optimal number of disks to have in the array? can you point me to any articles online comparing raid 5 performance with various numbers of disks? also are there any problems associated with having the system partition on the raid array or does it need to be on a seperate disk? thanks in advance for your help.

 

[SnoMunke]

You will probably not see a performance increase using 3 disks in a RAID 5 format...it actually may be slower than a single disk. However, what you with have is what RAID 5 provides -- fault tolerance.

Think of a single disk. Only pipe to pump data through.

Now think of how RAID 0 (striping) works. Now we have two or more drives to write to or read from...basically two or more pipes to pump data through...

Key: the more pipes the more data can be pumped through.

Now you have RAID 5 (striping w/ parity). With your three disks, you have two "pipes" to read or write through, but you also have to write parity to the third disk. This takes time and will lower any "gains" you might have gotten from writing (or reading from) the two disks.

So how do you increase performance? Remember the above key? That's right...add more disks. After adding "enough" disks, you will have enough pipes to pump data through to offset the time the system has to write to the parity disk.

There is a limit on how many disks you can use in a RAID 5 disk array before performance tops out. What that limit is I don't know...and your boss probably doesn't have the funds to find out (probably tens or hundreds of disks).

This pretty much sums my explanation. I would do a Google.com search on "RAID 5 performance" as do some reading from the Internet...the library at your fingertips...

later...

 

[dkozloski]

I'm running a RAID5 with four disks. I failed one disk and could harly tell the difference in performance when running on three. As previously stated the object of the RAID5 game is not performance but rather data integrity.

 

[dakarm]

10 U160 drives is about the max you want to put on a single RAID5 array.

Beyond that you wont see much of a performance gain, less then .5% for each drive added after the 10th.

Another thing you have to keep in mind is once a single drive fails, the array will be in a degraded state. Which mean everything a read or a write has to be peformed on any given stripe(drive) the parity information has to be recalculated or partity needs to be used to calculate the missing data.

This will slow down the performance dramatically.


I never create a RAID 5 with only 3 drives. My min. is 4 drives. 3 drives in the array and 1 drive as the spare.
Also you must keep doing tape backups. If that array fails (more then 1 drive = failed array) then you are SOL unless you have tape.

There is no formula for calculating RAID performance. I can make performance benchmarks look like it's flying along one minute then slower then IDE the next.

 

[bluefmc]

first, thank you all for your valuable input. perhaps i should have let you guys know what this server was going to be used for. this will be a linux web server serving both static html pages and pages generated by perl scripts. my understanding is that raid 5 is good for databases and such, but is it also a good choice for serving up webpages or would i be better off with a raid 10 (or another raid level) solution? and how many disks do you guys recommend? we have funds for up to 6 or so disks. and would it be better to have the system on a seperate disk from the data?

 

[SnoMunke]

At this point I think you should be talking to a Dell Engineer. Those guys should know the capability of their products. It is probably not a wise idea to be making business decisions based off peoples opinions in Anandtech forums. Not say nobody here knows what they are talking about, but it is probably best to talk to Dell. Otherwise, do some research from the Internet. Google.com baby! There is a wealth of technical knowledge out there I guarantee you won't find in these forums. (No offense to all forum members.) Good luck!

 

[dakarm]

Originally posted by: bluefmc
first, thank you all for your valuable input. perhaps i should have let you guys know what this server was going to be used for. this will be a linux web server serving both static html pages and pages generated by perl scripts. my understanding is that raid 5 is good for databases and such, but is it also a good choice for serving up webpages or would i be better off with a raid 10 (or another raid level) solution? and how many disks do you guys recommend? we have funds for up to 6 or so disks. and would it be better to have the system on a seperate disk from the data?

Don't no Never use RAID 5 for databases. When the array goes into a degraded mode it will cause all kinds of havic with the application. You always want to use RAID 0/1 for dtabases.


For what you want to use I would use RAID 5 but not use more then 5 diskn the array. That leaves1 disk for a spare. The best set up would be to set up a RAID1 for the OS and a RAID 5 for the data. <--- that's with MS OS. With linux you can probably get away with single drive for the OS and RAID5 for data. My question is how important is it for the server to be up. If you need to be up 99.99 percent I would go with RAID1 for OS and RAID5 for data regardless of the OS you use.