Getting the best out of raid.

[1stFalloutBoy]

For all of those out there getting into raid stuff, I have had many complaints over a period of time that peoples raids are under performing and I don't know if people actually understand why.

I'll quote an example using the Asus Striker II extreme , not to pick on Asus or this particular board as my understanding is that all boards operate in the way I'm about to describe.

Generally looking at a mother board you have 2 SATA connectors one above the other and manufacturers refer two this as two ports I.E SATA 1.0 and SATA 1.1, and they are ; however those two ports share bandwidth on one chip and thus if you plug an HD into each connection and stripe them the performance will pretty much suck.

Fixing the problem:
If you plug 1 HD into the Port SATA 1.0 and then plug another into SATA 2.0 and stripe it you will note quite a dramatic increase in speed as you are using the maximum bandwidth on each chip. I.E you should only plug one device in for each set of connectors in order to obtain maximum speed, this is correct also for PCIe SATA cards.

Personally I feel that when 1 Port will saturate the bandwidth of the chip running it that manufacturers should not refer to it as 2 SATA 3GBp/s ports as this is a little dishonest and will slow your drives down by using both connectors.

It also strikes me as very annoying that chipset manufacturers I.E intel are including so few SATA 3 connections on their motherboards... Whats the point in this, SATA 3 is AFAIK backward compatible with SATA 2, the only reason I can see that Intel is doing this is so that they can continue selling overpriced SAS/SATA controllers.

Whats your view.

 

[WyomingKnott]

For all of those out there getting into raid stuff, I have had many complaints over a period of time that peoples raids are under performing and I don't know if people actually understand why.

Because they don't know why they are using RAID, or what it actually does, so they use inappropriate setups. Yes, there are a lot of people out there who know what they are doing, but there are more who seem to think "RAID is good; I will put in some kind of RAID just because RAID is good."

However those two ports share bandwidth on one chip and thus if you plug an HD into each connection and stripe them the performance will pretty much suck.

In most cases, the chip and the bus connecting it to the core have plenty of bandwidth to support full transfer speed on both ports. By the way, if you only see two ports, you are using the Jmicron controller and not the six ports supported by the chipset.

Personally I feel that when 1 Port will saturate the bandwidth of the chip running it that manufacturers should not refer to it as 2 SATA 3GBp/s ports as this is a little dishonest and will slow your drives down by using both connectors.

I am willing to accept that you have tested this and found it to be true. It does, however, contradict everything that we think we know about chipsets. They are supposed to support full transfer speeds on all ports. The right way to test that is not with a RAID zero, but by running an independent drive on each channel and a test app that is capable of driving them all at once. Not seeing twice the throughput from two drives in RAID0 is just seeing an artifact of RAID0.

I

t also strikes me as very annoying that chipset manufacturers I.E intel are including so few SATA 3 connections on their motherboards... Whats the point in this, SATA 3 is AFAIK backward compatible with SATA 2, the only reason I can see that Intel is doing this is so that they can continue selling overpriced SAS/SATA controllers.

I don't understand. First of all, no rotating platter drive can use the full capacity of an SATA controller. There is no need for SATA 3 ports except for solid-state drives and some specialty hardware.


The board you mention is an LGA775 board. I think that it was released in 2006. There was no SATA III then, anyway.

If you plug 1 HD into the Port SATA 1.0 and then plug another into SATA 2.0 and stripe it you will note quite a dramatic increase in speed as you are using the maximum bandwidth on each chip.

Some questions. First, have you actually tried this? I'd love to see results. Second, I don't know what SATA 1.0 and SATA 2.0 are - is one a port on the Jmicron controller and one on the chipset? Third, in both tests that it sounds like you did (striping two drives on the same controller vs. striping two drives on different controllers), did you do the striping the same way? I assume that you had to do it in the latter case using Windows Disk Manager, since you can't RAID drives on different controllers via the BIOS. I want to know if you did the same for testing two drives on the same controller, because if you used RAID0 in that case then the tests aren't equivalent.

Much food for thought here. I'm torn between my experience telling me that you are off-base, and my willingness to look at concrete evidence.

 

[1stFalloutBoy]

For all of those out there getting into raid stuff, I have had many complaints over a period of time that peoples raids are under performing and I don't know if people actually understand why.

Because they don't know why they are using RAID, or what it actually does, so they use inappropriate setups. Yes, there are a lot of people out there who know what they are doing, but there are more who seem to think "RAID is good; I will put in some kind of RAID just because RAID is good."

While I agree with you about most people ... not in this instance, I personally was wanting to reduce load times, data security wasn't really an issue.

However those two ports share bandwidth on one chip and thus if you plug an HD into each connection and stripe them the performance will pretty much suck.

In most cases, the chip and the bus connecting it to the core have plenty of bandwidth to support full transfer speed on both ports. By the way, if you only see two ports, you are using the Jmicron controller and not the six ports supported by the chipset.

No I was only using the Striker II Extremes on board Sata II ports, I had 6 WD RE 1/2 Tb drives if I plugged in all 6 to the onboard top and bottom connectors the speed dropped significantly. When only 3 drives were connected on either the top or bottom but not both it was really fast, connect 2 drives Sata 1.0, 1.1, 2.0 and the speed sucked again 1.0,1.1 two drives speed sucked 1.0,2.0 speed was great.

For an example of what I'm talking about have a look at the Sata controller chip block diagrams, here
http://www.siliconimage.com/products/product.aspx?pid=63
this is for a Dual port Sata 150 controller but it holds true for this chip as well as I applied the same tests.

Personally I feel that when 1 Port will saturate the bandwidth of the chip running it that manufacturers should not refer to it as 2 SATA 3GBp/s ports as this is a little dishonest and will slow your drives down by using both connectors.

I am willing to accept that you have tested this and found it to be true. It does, however, contradict everything that we think we know about chipsets. They are supposed to support full transfer speeds on all ports. The right way to test that is not with a RAID zero, but by running an independent drive on each channel and a test app that is capable of driving them all at once. Not seeing twice the throughput from two drives in RAID0 is just seeing an artifact of RAID0.

No this one done too and it was noted that transferring files between 1 drive on 1 controller and 1 on another separate controller was fine , run two HD's on the ports attached to one controller and it tanks.

I

t also strikes me as very annoying that chipset manufacturers I.E intel are including so few SATA 3 connections on their motherboards... Whats the point in this, SATA 3 is AFAIK backward compatible with SATA 2, the only reason I can see that Intel is doing this is so that they can continue selling overpriced SAS/SATA controllers.

I don't understand. First of all, no rotating platter drive can use the full capacity of an SATA controller. There is no need for SATA 3 ports except for solid-state drives and some specialty hardware.

My point here is that SATA II is now a dated tech, SATA III being the latest spec gives you the full flat out speed, while I am using rotating platter drives ATM I may choose to replace every single drive with SSD's, why should I be stuck with old tech when it makes no difference to Sata 1 & 2 users as their devices will still work on a Sata III port?


The board you mention is an LGA775 board. I think that it was released in 2006. There was no SATA III then, anyway.

Yes I know at the time I was using Sata II drives but I believe looking at the latest Marvell controller that bandwidth is still shared with two drives and not when there is only one.

If you plug 1 HD into the Port SATA 1.0 and then plug another into SATA 2.0 and stripe it you will note quite a dramatic increase in speed as you are using the maximum bandwidth on each chip.

Yes.

Some questions. First, have you actually tried this? I'd love to see results. Second, I don't know what SATA 1.0 and SATA 2.0 are - is one a port on the Jmicron controller and one on the chipset? Third, in both tests that it sounds like you did (striping two drives on the same controller vs. striping two drives on different controllers), did you do the striping the same way? I assume that you had to do it in the latter case using Windows Disk Manager, since you can't RAID drives on different controllers via the BIOS. I want to know if you did the same for testing two drives on the same controller, because if you used RAID0 in that case then the tests aren't equivalent.

No the raiding was done via the raid controller Bios , this particular board ran an nVidia 790i chipset, windows was only involved when it came to formatting the drive for NTFS and had nothing to do with the creation of the Raid set.
The board has 8 sata connections in total 6 ports using the 790i and 2 on the JMicron, the raid testing in question was done solely on the chipset ports.

Think of it this way 1.0 is the connection above 1.1 but they are the same connector on the board I.E two headers in the connection.

1.0 | 2.0 | 3.0
1.1 | 2.1 | 3.1

Much food for thought here. I'm torn between my experience telling me that you are off-base, and my willingness to look at concrete evidence.

When I have my 2600K up and running ill run up the 1/2 TB's in the Q9450 again and do some bench marks for you... Any particular preference as to which benchmark to use, my only requirement is that it runs under Vista 64Bit ultimate.
Note - I cant speak for the current range of Intel chips as my 2600K is my first fully Intel setup for a while as my previous mobo's ran nVidias chipsets.

BTW Please excuse me not quoting your quote but I'm not that used to how the quoting of quotes in a quote works.

 

[WyomingKnott]

Judging from your response, you have done some serious testing and, as far as I can tell, tested the right things. So I believe that there are choke points in that chipset on that board. I am quite surprised.

I'm going to hazard a guess as to why current motherboards don't offer a collection of SATA III ports: The link from the SATA controller to the core has a certain limit, as you have shown, and they are trying to avoid exactly the issue that you describe. If there were six SATA III ports, the theoretical aggregate bandwidth would exceed the bandwidth between the controller and the core, and more people would complain as you did!

One other thought: having two drives on the same controller tank your throughput may be a controller design bug rather than a pipe width issue.

 

[1stFalloutBoy]

Judging from your response, you have done some serious testing and, as far as I can tell, tested the right things. So I believe that there are choke points in that chipset on that board. I am quite surprised.

I'm going to hazard a guess as to why current motherboards don't offer a collection of SATA III ports: The link from the SATA controller to the core has a certain limit, as you have shown, and they are trying to avoid exactly the issue that you describe. If there were six SATA III ports, the theoretical aggregate bandwidth would exceed the bandwidth between the controller and the core, and more people would complain as you did!

One other thought: having two drives on the same controller tank your throughput may be a controller design bug rather than a pipe width issue.

It is possible that this could be a controller design bug, it will be interesting if I have enough ports of the same type on the new motherboard to test things under the same conditions, I can tell you that on the 790i the difference in speed was worth the effort of all the testing, If the Intel chipset does not have these issues ill be a happy boy as it seems Z68 is somewhat limited in its number of ports and I am having to consider an add on sata III controller or a SAS/SATA config the sata III setup is cheap enough at 30NZD odd but will only in reality only give me one port the SAS/SATA setup would be nice but damn they are expensive starting at about 530NZD for a really basic model.

I wont be able to test this till Monday, I've got all my equipment just no case to put it in till then

 

[WyomingKnott]

What is this "case" of which you speak? Look at http://www.bluevalhalla.com/gallery/d/5754-2/IMG_0002.JPG

 

[1stFalloutBoy]

Unfortunately its not quite that easy for me, we have four cats a chihuahua and a Galah, not putting the unit into a case would result in it being eaten, knocked down, rubbed against and used as a scratching post.