RAID Optimization Guide

[maoten]

this is the best Raid Article i have read
Thanks alot for ur hardwork writing and testing
wish u Goodluck

[quux]

Originally Posted by RAST

1. RAID 0+1 does not have a “massive write penalty.” It as a 2 write penalty. You have to do 2 Write I/Os for each file system write. In other words when you save that word document each NTFS write I/O is doubled by the hardware. This makes sense because each side of the mirror has a full copy of the data.

Reading this you'd think that RAID1 takes twice as long to write as a single disk. Yes there are two writes, but they are simultaneous - they happen at the same time. So they don't take any longer than a single write.

I know you were talking about RAID0+1, but the point is that mirroring doubles the number of writes without doubling the amount of time those writes take. So, comparing a single disk to RAID1 mirrorset is a good conceptual exercise. Adding striping to the mirrorset just adds the speed advantages of striping to the redundancy advantages of mirroring.

[Adrian Wong]

Originally Posted by quux

Reading this you'd think that RAID1 takes twice as long to write as a single disk. Yes there are two writes, but they are simultaneous - they happen at the same time. So they don't take any longer than a single write.

I know you were talking about RAID0+1, but the point is that mirroring doubles the number of writes without doubling the amount of time those writes take. So, comparing a single disk to RAID1 mirrorset is a good conceptual exercise. Adding striping to the mirrorset just adds the speed advantages of striping to the redundancy advantages of mirroring.

That's true. But just wanted to clarify that it's only true if each hard disk is on its own IDE channel.

[quux]

True. But with an IDE RAID card, this is automatic - only one drive is used for each IDE interface on the card (well, that's what my 4-channel card said). Are there cards which let you use a master and slave on each channel? (I would stay away from such a thing.)

We discussed this in the virtual memory thread but it's worth mentioning in this thread too:

Personally, the Promise Fastrak100 is about the worst RAID card I've ever used. It is possible that basing one's experience of RAID on the use of this card would lead to a lot some bad conclusions about RAID overall (RAID1 slower, etc).

Also, there's a nice run-down of newer IDE RAID solutions over at Tweakers.net. Shows some progress since the days of the Fastrak 100!

[Adrian Wong]

Hmm.. If I'm not mistaken, you can run RAID even on a Master drive and a Slave drive on the same channel.

[pcmeiners]

"Reading this you'd think that RAID1 takes twice as long to write as a single disk. Yes there are two writes, but they are simultaneous - they happen at the same time. So they don't take any longer than a single write."

As to raid 1, the writes are NOT simultaneously written to both disks on any of the major hardware adapters I have used over the last 25 years.
In raid 1, the write completes on the first disk written to, then to the mirror. This design is implemented to inhibit corruption due to bad disk sectors on the first disk written to; if a bad sector corrupts data, as protection, it does not get written to the second disk. On the other hand, raid 10 writes are written simultaneously to more then 1 disk. Dependent upon the hardware cache and the setting thereof, the delay in raid 1 is basically eliminated if the cache is not flooded.

As a note to raid 1 or raid 10 stripe size...the stripe size of either can be changed without having to initialize ("reformat") which makes it very handy to find the "sweet spot" stripe size by testing with live data. You change the size and let a rebuild take place; you do not want to answer "yes" when prompted to initialize the disks as this will cause you you lose your data. Raid 5 does not offer this ability, if you change the stripe size the array is always initialized.

[Adrian Wong]

Originally Posted by pcmeiners

"Reading this you'd think that RAID1 takes twice as long to write as a single disk. Yes there are two writes, but they are simultaneous - they happen at the same time. So they don't take any longer than a single write."

As to raid 1, the writes are NOT simultaneously written to both disks on any of the major hardware adapters I have used over the last 25 years.
In raid 1, the write completes on the first disk written to, then to the mirror. This design is implemented to inhibit corruption due to bad disk sectors on the first disk written to; if a bad sector corrupts data, as protection, it does not get written to the second disk. On the other hand, raid 10 writes are written simultaneously to more then 1 disk. Dependent upon the hardware cache and the setting thereof, the delay in raid 1 is basically eliminated if the cache is not flooded.

As a note to raid 1 or raid 10 stripe size...the stripe size of either can be changed without having to initialize ("reformat") which makes it very handy to find the "sweet spot" stripe size by testing with live data. You change the size and let a rebuild take place; you do not want to answer "yes" when prompted to initialize the disks as this will cause you you lose your data. Raid 5 does not offer this ability, if you change the stripe size the array is always initialized.

Thanks for the heads-up, pcmeiners! We will update the guide accordingly.

[Adrian Wong]

SATA RAID seems to be the rage these days. Practically all motherboard manufacturers have at least a few motherboards with SATA RAID features built-in. With RAID support built into most motherboards and hard disk drive prices continuously dropping, more and more people are opting for RAID solutions as a cheap way to improve performance and/or reliability.

Today, we are going to take a look at how you can optimize your SATA RAID solution. Let's see what's the best RAID solution for your needs!

Link : RAID Optimization Guide Rev. 2.0