RAID 0 questions

[carlo75]

Hey I´ve just made an RAID 0 array whith my two seagate ST380013AS HD

Everything is working just fine, but I have some questions about RAID0 array.

1. Is there a performance gain when "disable tag queuing" option is enabled?
2. same question whith "disable synchronous transfers"??
3. MBM5 does not detect temperature sensors on my hard drives, neither does speedfan, how can I get then to display temperature?

thanks

[Gizmo]

I've not managed to get ANYTHING to display the temp sensors on my drives; I don't think that capability works with your average RAID controller.

I could be mistaken, but I believe the other two parameters are only relevant with SCSI controllers.

Command tag queuing allows the controller to reorder I/O requests to minimize head movement. This is good for servers as it allows them to maximize the number of I/O requests they can service, but not so good for desktop applications where the user wants something to happen Right Now. Generally, you would turn this off in a desktop environment.

Synchronous transfers are used in SCSI to increase data throughput. You NEVER want to disable this unless you are either a masochist or you have a really old SCSI device that you are hooking up (on second thought, strike that; you are definitely a masochist if you have a SCSI device that old). Basically, without synchronous transfers, the best data rate you will get is 10 MB/S, and that is best case; realistically you'll probably get closer to 3 MB/S. With synchrounous transfers, you can get burst rates as high as 320 MB/S, depending on the type of SCSI devices you are using and whether you have a PCI 133 MHz 64-bit SCSI controller.

[carlo75]

Originally Posted by gizmo

I've not managed to get ANYTHING to display the temp sensors on my drives; I don't think that capability works with your average RAID controller.

I could be mistaken, but I believe the other two parameters are only relevant with SCSI controllers.

Command tag queuing allows the controller to reorder I/O requests to minimize head movement. This is good for servers as it allows them to maximize the number of I/O requests they can service, but not so good for desktop applications where the user wants something to happen Right Now. Generally, you would turn this off in a desktop environment.

Synchronous transfers are used in SCSI to increase data throughput. You NEVER want to disable this unless you are either a masochist or you have a really old SCSI device that you are hooking up (on second thought, strike that; you are definitely a masochist if you have a SCSI device that old). Basically, without synchronous transfers, the best data rate you will get is 10 MB/S, and that is best case; realistically you'll probably get closer to 3 MB/S. With synchrounous transfers, you can get burst rates as high as 320 MB/S, depending on the type of SCSI devices you are using and whether you have a PCI 133 MHz 64-bit SCSI controller.

Ok got that!

So no performance increase on any of them? welll it´s clear tjat I wont turn of syncronous transfers...but as you said there is no point on having tag queuing enabled on desktop systems, as I understood...

thankx

[Gizmo]

I assume you are using the onboard IDE RAID. That being the case, I wouldn't expect them to affect your performance one way or the other.

[carlo75]

yes onboard raid on DFI infinity....

ok no performance increase, what about decrease?

[Áedán]

Command tagging can help on a server. However *IF* the drivers support it, you probably find it leads to a performance decrease. Most RAID drivers for SATA/IDE ignore the flag.

NCQ (Native command queuing) is something else though. The drives you have do support NCQ, but your controller might not. NCQ tends to increase performance a little bit.

[carlo75]

Originally Posted by Áedán

Command tagging can help on a server. However *IF* the drivers support it, you probably find it leads to a performance decrease. Most RAID drivers for SATA/IDE ignore the flag.

NCQ (Native command queuing) is something else though. The drives you have do support NCQ, but your controller might not. NCQ tends to increase performance a little bit.

So if I check that option (disable tagging) it would be performance decrease?

[Áedán]

Only if the SATA controller and the drivers you have can actually support NCQ. If they don't support NCQ, then it's fairly pointless the drives having it.

With NCQ, all three components (drivers, SATA controller and hard disk) have to support it for NCQ to be usable. If any one of the components doesn't support NCQ, then you can't use it.

[carlo75]

Originally Posted by Áedán

Only if the SATA controller and the drivers you have can actually support NCQ. If they don't support NCQ, then it's fairly pointless the drives having it.

With NCQ, all three components (drivers, SATA controller and hard disk) have to support it for NCQ to be usable. If any one of the components doesn't support NCQ, then you can't use it.

how can I know if they can support it? my HD are ST380013AS seagate, and I use the SATA raid on my DFI infinity Silicon inmage sil3114 PCI to serial ATA controller

[cloasters]

Let's see if I have this right: RAID 0 keeps your data safe, while RAID 1 decreases the time needed to grab the desired data from your HDD's. Is this a true story?

Pardon this second question, please. What does RAID 0 + 1 do? Must one have three HDD's to run 0+1?

[Gizmo]

Eh, backwards Cloasters.

RAID0 is the fastest RAID there is, especially with several drives. Unfortunately, it is also begging for trouble, as it isn't truly RAID (the R in RAID means Redundant, and RAID0 is NOT redundant).

RAID1 is the most reliable, because you have two complete copies of all your data; one on each drive. It's also pretty darned fast for reads, because you can stagger reads across the two drives in a RAID, but slower for writes because all of the data has to be written twice (once to each drive). It also has the drawback that you can only build a RAID1 with two drives, and you throw away have your total drive space on redundancy. The slower write issue is actually a wash for RAID configurations because ANY RAID is going to require two writes (once for the real data and once for the redundant data).

RAID 0+1 is an attempt to combine the benefits of RAID0 with RAID1. You build a RAID0 array with some number of drives. You then build another RAID0 array just like your first one, and you make the controller mirror the first RAID0 onto the second as a RAID1 (hence, the RAID 0+1 moniker). This allows you to achieve read performance similar to RAID0, but have the data protection RAID1 offers. Unfortunately, it also suffers from RAID1's biggest drawback: You throw away half your storage.

There is also a RAID 1+0, which takes multiple RAID1 arrays and stripes them together in a RAID0. In a RAID 0+1, you can actually tolerate the failure of up to half the drives in the array, as long as all the drives that fail are in the same RAID0. Random chance, though, says that half your failures are going to be in each of the RAID0 arrays that comprise the RAID1. This means that if you lose two drives, you are probably toast. RAID 1+0 attempts to address this. Since this configuration is multiple RAID1s striped together as a RAID0, you reduce the odds of a two drive failure taking down the entire RAID. Statistically, if you have x RAID1 arrays, then then the odds are that you will have to have at least x+1 drive failures before your entire RAID is toast. However, as with RAID 0+1, you still lose half your storage on redundancy.

RAID3 is another effort to get some of the benefit of RAID0 without the drawbacks of RAID1 (losing half your storage). With RAID3, the data are striped across the drives in an array, just like RAID0, but one of the drives in the array is dedicated for parity (redundancy) info. Because of the parity calculation, RAID3 is redundant in the same way that RAID1 is redundant, but you only throw away the equivalent of one drive for redundancy, instead of half your storage. In addition, you can have an arbitrary number of drives. This means that a RAID3 with 5 drives is only using 20% of its storage for redundancy, instead of 50% like RAID1, RAID 0+1, or RAID 1+0. However, RAID3 has one serious drawback. For small data writes, it stinks, because of the dedicated parity drive. If your data are smaller than the stripe size (usually 64 KB) and you make multipe small data writes, the odds are that you are going to end up writing all of your data to different drives in the RAID (particularly as you increase the number of drives in the RAID). However, all of the parity info gets written to a SINGLE drive, so your RAID write performance ends up being limited to the rate that you can write data to that parity drive.

RAID5 is an attempt to address the weakness of RAID3. It changes the parity arrangement so that parity 'floats' across all of the drives. This 'floating' is complex, but it gurantees that the data are never written to the same drive that carries the parity for that data, and also guarantees that no single drive carries all of the parity. Thus, the performance bottleneck is removed. RAID5 is a bit slower on writes than RAID1 because of the parity calculation, but for reads it rivals RAID0, and it can be done with an arbitrary number of drives. In addition, if you use a dedicated hardware controller for RAID, the performance overhead of the parity calculation is negligible. RAID5 shines best on database systems or large storage arrays with a lot of random access, IMHO, particularly if implemented with a large number of drives.

Most dedicated controllers also support either RAID5+0 (RAID50) or RAID 5+1 (RAID51). IMHO, RAID50 is a waste of time for all but the most severe speed freaks, and RAID51 is a waste for all but the most paranoid.

Now you know more about RAID than you ever wanted to, your eyes have probably glazed over and you are developing a severe migraine from information overload.

[cloasters]

Thanks Gizmo! That's a lot of information about RAID. I guess a RAID 5 set up is needed for a PC without critical information on it. Might as well make it from 15k rpm 72 Gb SCSI 320 HDD's, too.

Once upon a time I was almost that crazy!

[mookydooky]

RAID5 is definately the way to go. If you're super-paranoid about your data though, you can do RAID6, which inserts another parity set. RAID6 will allow you to lose 2 drives and still be able to rebuild the data.

[Gizmo]

True. I've never seen a controller with RAID6, though, so I can only assume that it is only available on controllers costing better than $1000. There's also a RAID4, which differs from RAID3 in that RAID3 stripes at the byte level and RAID4 stripes at the block level (basically, RAID3 uses a much smaller stripe size than RAID4, when you get down to it). I've only ever seen one controller that implemented RAID4 (which was called RAID3 by the manufacturer) and none that implemented true RAID3.

I understand there is also a RAID7, although the definition of RAID7 seems to differ depending on who you talk to.