Does Raid 0 actually improve performance?

[Pariah]

First off, SR did not write that, they borrowed it from another source. SR's own opinion posted in the link that Gosharks posted is in stark contrast to the above:

"Finally, looking at the bigger picture, it's often better to use the two identical drives individually rather than in a RAID config. This is because two (or more) independant processes/applications trying to access the drive or array at once can cause the heads to be seeking back and forth trying to fulfill both requirements, crippling performance. Depending on the situation, it's possible that two drives could be serving data to the two processes without all the seeking, thus improving performance significantly."

So, now the question, is what Gosharks has stated, what determines a smart controller and who makes them? The guy who wrote the RAID faq is talking in theoretic terms rather than actual examples. I think we agree with Matthias that this is possible on truly enterprise class hardware like SAN's where havine 1000 drives stuck waiting for one file would be death, however, I have never seen this referred to anywhere when talking about hardware any of us could afford and install in our computers.

 

[smahoney]

First - I am a senior storage engineer and have been working with RAID systems for more than 10 years. That being said there is a vast difference between most of the RAID controllers that are available for home users and those that are installed in large servers or standalone RAID subsystems or SAN attached arrays such as the EMC CLARiiON or Symmetrix lines. RAID does offer performance benefits over single drives in nearly all situation, but there are exceptions. The 'average' home user will not experience a noticeable benefit of utilizing a RAID 0 array for their OS drive compared with a single fast drive such as a WD Raptor. It is how an application behaves that dictates any performance benefits of RAID. Often, using separate disks for different function is more efficient than RAID since it reduces contention. Examples would include placing your paging file on a separate drive, a separate drive for backup, video streaming, MP3's. The type of RAID used also has an impact on application performance. Servers running SQL or Exchange have very different performance demands than your typical home user. Several review that compare SCSI and SATA drives show how they are tuned differently for different applications. WD 74GB Raptors do very well against even 15K SCSI drives in business benchmarks and games but drop quickly in server benchmarks.

'Smart' RAID controllers usually refer to those that have additional cache, battery backup, hot swapping support, dedicated XOR engines for RAID 5 or external arrays that may have multiple controllers and failover.

Home users need to remember that using a RAID 0 array will not improve the response time of their hard drives, it will usually only show a difference when reading or writing large files. RAID 1 often imposes a slight performance hit that is usually not noticeable by the average user. Games generally show little or no improvement. Often simple hard disk management can provide for better performance increases - things like periodic defragmentation and smarter layouts of partitions. Considering the availability of relatively inexpensive memory, RAM disks using inexpensive utilities are more and more common.

I admit that I use a RAID 0 array for my OS, by I also short stroked the partitions and place my archive data on a separate disk with a backup image and all my driver files.

The other issue I have is that most of the benchmarks mentioned in previous posts that look at more than two drives are using controllers that are not the most stable. A good quality RAID controller for RAID 5 or RAID 10 costs real money. Don't be surprised if you perform a BIOS update on an inexpensive controller and find out you just corrupted your 1TB RAID 5 array of 4 250GB drives. Some of the SATA controllers aren't bad, but they lack many of the monitoring tools and reliability of the more expensive SCSI types.

I would also not recommend setting up a RAID 0 array of more then 2 drives unless you really don't care if you lose the data. You should of course backup your vital data on a regular basis not matter what type of RAID array you are or are not using.

PCI bandwidth can also be an issue - Server tend to be running 64 bit controllers in PCI-X at 66MHz or better - roughly 4 or more times the performance of a home PC PCI slot which can reduce the maximum effective bandwidth of a large array - although usually only in burst mode.

I would still like to try out a Promise S150SX4 with 256MB of cache and 4 WD740GD's - all for less than $1200.

 

[Markfw]

Originally posted by: Regs
Mark, how much is your CPU utilization of the SCSI Raid set up?

Which benchmark should I try ? HD tach says about 40%, but I really don't believe that.ATTO and Sandra don't have that number.

 

[T9D]

..

 

[Markfw]

Originally posted by: smahoney
First - I am a senior storage engineer and have been working with RAID systems for more than 10 years. That being said there is a vast difference between most of the RAID controllers that are available for home users and those that are installed in large servers or standalone RAID subsystems or SAN attached arrays such as the EMC CLARiiON or Symmetrix lines. RAID does offer performance benefits over single drives in nearly all situation, but there are exceptions. The 'average' home user will not experience a noticeable benefit of utilizing a RAID 0 array for their OS drive compared with a single fast drive such as a WD Raptor. It is how an application behaves that dictates any performance benefits of RAID. Often, using separate disks for different function is more efficient than RAID since it reduces contention. Examples would include placing your paging file on a separate drive, a separate drive for backup, video streaming, MP3's. The type of RAID used also has an impact on application performance. Servers running SQL or Exchange have very different performance demands than your typical home user. Several review that compare SCSI and SATA drives show how they are tuned differently for different applications. WD 74GB Raptors do very well against even 15K SCSI drives in business benchmarks and games but drop quickly in server benchmarks.

'Smart' RAID controllers usually refer to those that have additional cache, battery backup, hot swapping support, dedicated XOR engines for RAID 5 or external arrays that may have multiple controllers and failover.

Home users need to remember that using a RAID 0 array will not improve the response time of their hard drives, it will usually only show a difference when reading or writing large files. RAID 1 often imposes a slight performance hit that is usually not noticeable by the average user. Games generally show little or no improvement. Often simple hard disk management can provide for better performance increases - things like periodic defragmentation and smarter layouts of partitions. Considering the availability of relatively inexpensive memory, RAM disks using inexpensive utilities are more and more common.

I admit that I use a RAID 0 array for my OS, by I also short stroked the partitions and place my archive data on a separate disk with a backup image and all my driver files.

The other issue I have is that most of the benchmarks mentioned in previous posts that look at more than two drives are using controllers that are not the most stable. A good quality RAID controller for RAID 5 or RAID 10 costs real money. Don't be surprised if you perform a BIOS update on an inexpensive controller and find out you just corrupted your 1TB RAID 5 array of 4 250GB drives. Some of the SATA controllers aren't bad, but they lack many of the monitoring tools and reliability of the more expensive SCSI types.

I would also not recommend setting up a RAID 0 array of more then 2 drives unless you really don't care if you lose the data. You should of course backup your vital data on a regular basis not matter what type of RAID array you are or are not using.

PCI bandwidth can also be an issue - Server tend to be running 64 bit controllers in PCI-X at 66MHz or better - roughly 4 or more times the performance of a home PC PCI slot which can reduce the maximum effective bandwidth of a large array - although usually only in burst mode.

I would still like to try out a Promise S150SX4 with 256MB of cache and 4 WD740GD's - all for less than $1200.

Could you comment on an LSI logic Megaraid 1650 (wih 128 meg cache) ? With five Seagate Cheetah (ST318404LW) 10k SCSI in raid5 (or raid0, tried both) with a 64k stripe size ? This is in regards to your comment on "controllers mentioned above with more than 2 drives" as shuttleteam (a well known SCSI expert here) recommended this as one of the best in it class, small servers.

 

[Regs]

I do admit I seem some what convinced of the performance you see. Yet I would still be some what hesitant to spend that much cash on a RAID set up. For workstations and servers I think of it as almost a necessity but for a desk top it's very pricy.

I play games and design web pages. And for Photoshop and flash, all I really need is a skuzzy or SATA drive like I have now. The Raptor is more than plenty to handle my 2D files. 3D on the other hand is a different story, but most 3D work is down on a workstation computer which is none of my concern thankfully.

It still comes down to the fundamental point that RAID is not for every one.

 

[Markfw]

Originally posted by: Regs
I do admit I seem some what convinced of the performance you see. Yet I would still be some what hesitant to spend that much cash on a RAID set up. For workstations and servers I think of it as almost a necessity but for a desk top it's very pricy.

I play games and design web pages. And for Photoshop and flash, all I really need is a skuzzy or SATA drive like I have now. The Raptor is more than plenty to handle my 2D files. 3D on the other hand is a different story, but most 3D work is down on a workstation computer which is none of my concern thankfully.

It still comes down to the fundamental point that RAID is not for every one.

and you know what ? If I had it to do all over again, I wouldn;t have spent $650 on that array. 2 Raptors in raid 0 is what I would have done.
(20/20 hindsight is easy) But its fun now, and great to enjoy the happiness if true SCSI speed !!

 

[MichaelD]

and you know what ? If I had it to do all over again, I wouldn;t have spent $650 on that array. 2 Raptors in raid 0 is what I would have done

Heheheh, yeah, it does get expensive, doesn't it?

I can't really wade-in on the technical stuff you guys are arguing about, but I'll throw my two cents in.

I've been running RAID-0 (which isn't really RAID, technically...) for years now. And as has been said, It's not for everyone. Setup requires a little more effort and thought than a single-drive install does.

But I've never had an array fail, I do back up my system (incremental backup) 1x/week and really important stuff is saved on CDR as well.

I do notice a big difference in seat-of-the-pants speed. In particular, the OS loads faster, games/levels load faster. That's enough for me. I like it and I'm keeping it.

In the future, as SATA spreads and prices come down, I'll probably go with two 74GB drives, striped. The $340 I spent on 36GB of SCSI storage for the array I'm running now was pretty steep...but it's something I wanted to do. I wanted to get my feet wet w/SCSI and test it out.

I've never personally used SATA drives, but SCSI blows IDE out of the water when doing anything besides running a single Application. I.E. I can encode MP3s, while surfing, while listening to music all while NAV scan runs in the background w/almost no discernable diff in system "snappiness."

FWIW.