Raid 0,1 and 5

Please help me understand about the the differences between raid 1 and 5 since I am not too clear about how this work. From what I understand, when you have raid 5 and you have 5 physical disks, it will stripe to 1 big disk so if you write to the disk it will write randomly to one or two out of these 5 disks right ??? Here is what we do in my company and I think it's not right ??? they have raid 5 and they only have one partition since they said no need to have many different partitions since i will write to different disks any way, it only cause overhead if we have more than one partition. Is that a true statement ??? So we have only one partition and they put datafiles and index files in one partition. Is that right ???

the system (C drive) and (E drive) will be partitioned on pair of mirrowed and they put all of the log files and system, rollback in there.

to me this configuration is not right but I like to prove it to them.

Thanks

Quote:

---

Originally posted by mike73
Please help me understand about the the differences between raid 1 and 5 since I am not too clear about how this work.

---

There's lots of information at http://dbasupport.com/oracle/resourc...Hardware/RAID/ about RAID configurations. Basically, RAID 1 is mirroring. For every disk, you have another disk.

RAID 5 uses what's called parity striping. With RAID 5, each write is spread out over all the physical disks along with parity information. If one of the disks is lost, you can continue at a diminished capacity. If you lose two disks, you're SOL.

Quote:

---

From what I understand, when you have raid 5 and you have 5 physical disks, it will stripe to 1 big disk so if you write to the disk it will write randomly to one or two out of these 5 disks right ???

---

No, not quite. The writes go to all disks in the RAID volume.

Quote:

---

Here is what we do in my company and I think it's not right ??? they have raid 5 and they only have one partition since they said no need to have many different partitions since i will write to different disks any way, it only cause overhead if we have more than one partition. Is that a true statement ???

---

You get no performance advantage by having multiple partitions on the same RAID volume.

Quote:

---

So we have only one partition and they put datafiles and index files in one partition. Is that right ???

---

In the ideal world, I wouldn't do it like that. You are setting yourself up for I/O contention.

Quote:

---

the system (C drive) and (E drive) will be partitioned on pair of mirrowed and they put all of the log files and system, rollback in there.

to me this configuration is not right but I like to prove it to them.

Thanks

---

Having redo, rollback, temp, and system on mirrored devices is fine.

Hi,
The difference between raid 1 and raid 5 is simply that raid 5 comes with ECC that stripes your data in in 5 physical drives even though you are looking at it as 1 logical drive; and Raid 1 does stripe your data with only one parity.
Files that are accessed sequentially as Temp tablespace and redo log should never be placed on raid 5.

Check the documentation Jeff refers you to.

Thanks
Richard

Quote:

---

Originally posted by richardana

Files that are accessed sequentially as Temp tablespace and redo log should never be placed on raid 5.

---

I'm ALWAYS careful with the words "ALWAYS" and "NEVER".

Redo logs are better on RAID 5 devices than on unprotected devices even though your performance will be degraded.

I thought redo logs is one of those files types that can exist on non-redundant disks, as long as you mirror properly in Oracle. Am I wrong?

and here is a great link on RAID types:
http://www.acnc.com/04_01_0p1.html

Quote:

---

Originally posted by richardana

Raid 1 does stripe your data

Files that are accessed sequentially as Temp tablespace

---

RAID 1 is not stripping, it is mirroring, RAID 0 is stripping

Temp tablespace is not access sequentially, redo logs are but not tablespaces

Jeff
You 're right about the use of the word NEVER. What I meant to say was " Files that are sequentially accessed SHOULD NOT BE placed on raid 5.
Thanks
Richard

Jeff,

I am a little confused per your email. On your reply, one paragraph said it's ok to have one partion and the next paragraph said if I put data files and index files in the same partition, I will set myself in disk I/O. What do you mean by saying that ??? Can you give some more detail ?????

Thanks

In your case, putting all of your disks in one RAID volume will probably not help performance. Ideally, you should have multiple volumes that contain multiple disks in order to maximzie throughput. For example, with your 5 disks, you could have a 3 disk RAID 5 volume and a 2 disk RAID 1 volume. This way, you could spread your I/O over multiple spindles but still limit your I/O contention by placing some files on the RAID 5 volume and some other files on the RAID 1 volume.

IMHO, you need at least 12 disks over 2 controllers to have decent performance in an Oracle Database.

Quote:

---

Originally posted by Zaggy
as long as you mirror properly in Oracle.

---

Yes, this is the key. As long as you multiplex your files, you can use them on unprotected devices. If you don't multiplex, I would rather have them on a RAID 5 than an unprotected RAID 0.

Quote:

---

As long as you multiplex your files, you can use them on unprotected devices. If you don't multiplex, I would rather have them on a RAID 5 than an unprotected RAID 0.

---

Right.
I recommend that you use RAID 5 only if you run DSS, a read only Orcale DB (or a system which reads data most of the time).

I use RAID 1+0. It is faster than RAID 5 bacause it has less computation overhead.

Now, you may ask what is the problem with writing in RAID 5. As you probably know fisrt data stripe is read from the array, next the new data is added to the stripe, next you have to compute the new parity block (there are several algorithms for that) and eventually the data is written to the disk.

What I have heard is that after all RAID 5 might decrease writing by 20-30% but it might increase reading (big reads).

I agree with Julian also, I was working on a system with a high insert load, around 700 row inserts per second 24x7, the server was originally set up with RAID 5, but because of the Parity building/checking, it was killing performance. I ended up going straight RAID 0 stripping over 6 drives for the fastest speed. I was running two servers in parallel for failover, therefore morroring and parity wasn't required.

Maybe RAID 0 + 1 (Stripping + Mirroring) would be a good option if you can afford the storage space.