6_6_6

NCQ: Best Upgrade For a Power User!

Recommended Posts

So I'm very confused because I couldn't find ANYWHERE else , a review, a benchmark , that would confirm what I read on this

thread.

I gave the links to the first benchmark thread done by SOMEBODY ELSE on this forum. This thread confirmed CZR's his findings.

After reading this thread, I'm just lost ! So, Please advise !

Well, reread the thread. I answered it many times. I don't understand how people are getting lost with performance when only one drive series performs 80 mbps throughput on 10 concurrent reads and all the rest drop down to 20-30 mbps.

Edited by 6_6_6

Share this post


Link to post
Share on other sites
@Linux: perhaps Linux does something comparable to NCQ itself, like the 3ware RAID controller of my mate. That would explain everything.

Linux is reordering disk accesses for optimal order since decades.

Widows probably too.

My guess is, that in case of NCQ Windows thinks "Oh, this drive can order operations itself, so I won't bother" and sends request as they are, but then shitty CQ algorithms in the drive firmware screw up everything. Remember, while CQ works in SCSI hard drives for decades, in SATA it is a new thing and has yet to get through it's childhood diseases.

Simply put: PROPER WORKING NCQ is without a doubt, the single most important upgrade for a power user in almost a decade i believe. Now you can multitask.

Or you simply continue without NCQ and have the (more or less) same performance ;)

Share this post


Link to post
Share on other sites
Linux is reordering disk accesses for optimal order since decades.

Widows probably too.

My guess is, that in case of NCQ Windows thinks "Oh, this drive can order operations itself, so I won't bother" and sends request as they are, but then shitty CQ algorithms in the drive firmware screw up everything. Remember, while CQ works in SCSI hard drives for decades, in SATA it is a new thing and has yet to get through it's childhood diseases.

I don't think so. That would mean that if you have a drive without NCQ or you have NCQ disabled (by putting the controller into IDE/legacy/whatever mode), you would have a much higher throughput with more than one program accessing the drive at the same time, because Windows would do the command reordering. But that does not happen.

Or you simply continue without NCQ and have the (more or less) same performance ;)

Why that?? The benchmark results and 6_6_6 reports suggest that the performance is not the same at all! :blink:

@extrabigmehdi: today I get my two 7200.11 640GB. During the next days I'll post how good or bad they are. :)

If you believe the benchmarks from the other guys, there is only one hard drive series you should consider to buy: Seagate 7200.11.

Edited by FAT_Punisher

Share this post


Link to post
Share on other sites

Simply put: PROPER WORKING NCQ is without a doubt, the single most important upgrade for a power user in almost a decade i believe. Now you can multitask.

Or you simply continue without NCQ and have the (more or less) same performance
;)

xerces8, how is that so?

These are your benchmarks:

Disk 0 : Western Digital WD7500AAKS

1 instance : 98 MB/s

2 instances : 14 MB/s

9 instances : 21 MB/s

Disk 1 : Samsung HD642JJ

1 instance : 116 MB/s

2 instances : 65 MB/s

9 instances : 21 MB/s

And this is 7200.11:

SEAGATE ST3750330AS 7200.11 (750 GB, 7200 RPM, SATA-II)

1 instance: 110 MB/s

2 instance: 103 MB/s

10 instances: 75-80 MB/s

80 MB/s: My 7200.11 with 10 instances (NCQ) [/color]

21 MB/s: Your WD7500AAKS with 10 instances (No NCQ)

102 MB/s: My 7200.11 with 2 instances (NCQ)

14 MB/s: Your WD7500AAKS with 2 instances (No NCQ)

Please revise / edit your post. You will be confusing people some more.

Edited by 6_6_6

Share this post


Link to post
Share on other sites

I think I must puke...

SEAGATE ST3640323AS 7200.11 (640 GB, 7200 RPM, SATA-II)

1 instance: ~110 MB/s

2 instances: ~53 MB/s (avg. 25,3 + 27,3) (0% and 50%, 256K blocks)

4 instances: ~40 MB/s (0%, 33%, 66%, 99%)

Measured on the Dell Perc 5/i. The Intel ICH9R delivers even shittier results, just like it does with the WDs.

It's a bit faster than a RAID 0 of two WD6400AAKS, but still quite shitty.

I see no big difference between the WD6400AAKS and the Seagate ST3640323AS. :angry:

Edited by FAT_Punisher

Share this post


Link to post
Share on other sites

Not good. Don't lose hope yet. Let's troubleshoot for you. I can assist with ICH9R. That is what i have.

1. Is AHCI is turned on in BIOS?

2. Is Intel Matrix Storage Manager is installed?

3. Are you watching the bandwidth from TASK MANAGER > PERFORMANCE > RESOURCE MONITOR > DISK?

This looks like non-ICQ for me. This is not part of some RAID. right? Single disk installation.

I think I must puke...

SEAGATE ST3640323AS 7200.11 (640 GB, 7200 RPM, SATA-II)

1 instance: ~110 MB/s

2 instances: ~53 MB/s (avg. 25,3 + 27,3) (0% and 50%, 256K blocks)

4 instances: ~40 MB/s (0%, 33%, 66%, 99%)

Measured on the Dell Perc 5/i. The Intel ICH9R delivers even shittier results, just like it does with the WDs.

It's a bit faster than a RAID 0 of two WD6400AAKS, but still quite shitty.

I see no big difference between the WD6400AAKS and the Seagate ST3640323AS. :angry:

Share this post


Link to post
Share on other sites

I removed a tiny jumper from the Seagate.

At the Intel ICH9R, it now shows the shitty jumping behaviour the WDs had shown as well (alternating randomly between 0 B/s and full-speed). And no, that's not a desireable behaviour, on the contrary, it's the reason why I came here in the first place, because my apps froze and didn't respond for a time (probably that time in which HD_Speed shows 0 B/s...). The Perc 5/i solved that. Throughput is shitty, but at least it's distributed equally among all apps.

So for me that's it. I'll send back one Seagate and sell the other.

Perhaps I will transfer my OS to the Seagate to see how it "feels". Because what I "felt" was the reason for my research in the first place. I don't care about numbers.

Edited by FAT_Punisher

Share this post


Link to post
Share on other sites

These are quantifiable numbers. They do matter.

First make sure a single drive works as it should with ICH9R. Then you can go about your RAID. I have no idea what Perc 5/i is.

Make sure it is AHCI in BIOS and not RAID. Make sure it is Intel Storage Manager and not Microsoft. Make sure you are not plugging the drive to an onboard jmicron controller. Then try again.

2 instances you must hit 100 MB/s (at 0 and 50% positions) if it is working for you (in DISK MONITOR). If it is not, don't bother testing 10 instances until you make it so.

I removed a tiny jumper from the Seagate.

At the Intel ICH9R, it now shows the shitty jumping behaviour the WDs had shown as well (alternating randomly between 0 B/s and full-speed). And no, that's not a desireable behaviour, on the contrary, it's the reason why I came here in the first place, because my apps froze and didn't respond for a time. The Perc 5/i solved that. Throughput is shitty, but at least it's distributed equally among all apps.

So for me that's it. I'll send back one Seagate and sell the other.

Perhaps I will transfer my OS to the Seagate to see how it "feels". Because what I "felt" was the reason for my research in the first place. I don't care about numbers.

Share this post


Link to post
Share on other sites

Well, actually, that is how it is and it should be under NCQ. So NCQ is probably working for you. Check my previous graphs.

Here it is for a properly working NCQ: http://forums.storagereview.net/index.php?...st&p=251948

I removed a tiny jumper from the Seagate.

At the Intel ICH9R, it now shows the shitty jumping behaviour the WDs had shown as well (alternating randomly between 0 B/s and full-speed). And no, that's not a desireable behaviour,

Edited by 6_6_6

Share this post


Link to post
Share on other sites

@6_6_6

I gave the links to the first benchmark thread done by SOMEBODY ELSE on this forum. This thread confirmed CZR's his findings.

I'm very sorry, but everything remain in this website.

No other "well known" website mention this.

Just wondering if people are not working for seagate , there.

And why seagate wouldn't advertise this, if this would give them a clear advantage ?

Well, reread the thread. I answered it many times. I don't understand how people are getting lost with performance when only one drive series performs 80 mbps throughput on 10 concurrent reads and all the rest drop down to 20-30 mbps.

I still hesitate because:

* not everyone seems to configure properly a SATA 2 connection, and hence enable NCQ

* some people don't understand they must give the total throughput, and not average.

* I need more confirmations by other users. One success, doesn't mean same for others.

* IOMeter tests seems to contradict what is said here.

* velociraptor hasn't be tested, yet for it's NCQ ability. You can't compare Velociraptor to Raptor,

because the Velociraptor have a SATA2 connection , the Raptor not.

* given the high failure rate of seagates, I still might buy a shitty drive,

andd then still NCQ at all.

By the way, I'm waiting for FAT_Punisher final results.

I might at the end , buy a velociraptor AND a seagate 7200.11. It just seems a bit like a waste. (of time, money).

Edited by extrabigmehdi

Share this post


Link to post
Share on other sites
These are quantifiable numbers. They do matter.

They do not matter if an application freezes, if a running TV recording comes to a stop with the movie incomplete and corrupted, because the application could not write to the disc for a too long time. In a situation like that I do not care about numbers at all.

With the Intel ICH9R, I have situations like that. With the Perc 5/i, I do not have them. With my old nVidia nForce4 onboard RAID controller, I did not have them either.

First make sure a single drive works as it should with ICH9R. Then you can go about your RAID. I have no idea what Perc 5/i is.

I have tested a single drive.

The Perc 5/i is a professional SATA/SAS RAID controller which I mentioned several times in this thread, so this basically says that you didn't read many of my posts. :(

Share this post


Link to post
Share on other sites

Listen dude. Nobody is forcing you to buy what you want to buy. It is not as if i am getting commission out of your pocket. I got my stuff working months ago as i wanted. If i am wasting my bloody time here, it is for you people. It doesn't benefit me whether your run seagate velociraptor or a rocket launchpad.

We are all enthusiasts here. If you want to contribute, do some tests. If not, go to some other "well known" site whatever that is and ask them to run your tests. But do not ridicule our efforts.

@6_6_6
I gave the links to the first benchmark thread done by SOMEBODY ELSE on this forum. This thread confirmed CZR's his findings.

I'm very sorry, but everything remain in this website.

No other "well known" website mention this.

Share this post


Link to post
Share on other sites
These are quantifiable numbers. They do matter.

They do not matter if an application freezes, if a running TV recording comes to a stop with the movie incomplete and corrupted, because the application could not write to the disc for a too long time. In a situation like that I do not care about numbers at all.

You are talking about something else FAT_Punisher. If you have that, you probably have bad sectors or weak sectors on your hard drive. Get it checked by mhdd or hddscan first.

The Perc 5/i is a professional SATA/SAS RAID controller which I mentioned several times in this thread, so this basically says that you didn't read many of my posts. :(

I read that. It means I don't have an experience with that. First you make it work with ICH9R, then you can try other solutions.

Now, from your posts, it looks like NCQ is working. Please answer my questions so that we can work on this.

Share this post


Link to post
Share on other sites
You are talking about something else FAT_Punisher. If you have that, you probably have bad sectors or weak sectors on your hard drive. Get it checked by mhdd or hddscan first.

Seriously... :rolleyes:

I have checked for bad sectors, and the phenomenon occured on two different (!) WD6400AAKS. It's not a bad sector thing, especially since using a different RAID controller (the Perc 5/i) solves the problem!

Now, from your posts, it looks like NCQ is working. Please answer my questions so that we can work on this.

The controller is running in RAID mode, but the Seagate is not part of a RAID.

It looks like those benchmarks you did yourself (alternating between 0 and full-speed, even with 9 HD_Speeds), so you would call it working, yes.

But.

I've just made a real-world test, not that HD_Speed stuff.

I copied a huge file from the Seagate to a RAID 0 (which can write it twice as fast as the Seagate can read it) and checked an image at the same time, and adding to that played a video, all on the Seagate. So I had two apps that are able to read with full-speed if run alone (and they do, when I stop one of them).

The total reading throughput was only 20-30 MB/s, which very clearly shows that there is no real NCQ happening!

Bottom line: the 7200.11 can't do NCQ better than the WDs. And if this is really working NCQ, it's pointless. No huge multitasking gain at all.

Share this post


Link to post
Share on other sites
The controller is running in RAID mode, but the Seagate is not part of a RAID.

Seagate is part of what then? You cannot turn AHCI/RAID on or off per drive in BIOS.

All my tests has been done in AHCI. And with Intel Matrix Storage manager.

If you want to test this stuff properly, please make sure you can mimic our scenario first then confirm if it works or not. Then you can try your RAID / Perc or other special requirements (once you make sure it works).

And now, forget about copying / transfering or manually trying stuff. Believe me, if it is working it should, you must have no drop in performance in 2 instances of hddspeed. Once you have that, you can manually test real world scenarios.

Share this post


Link to post
Share on other sites

hi Fat punisher,

could you download & install "hd tune".

Now please check in the info tab, that NCQ is properly activated, like in the picture:

sata2-ncq.png

Also go to the health tab , and see if global health satut is ok.

Especially check "reallocated sector count" statut. Ideally, the line is white.

test.jpg

Share this post


Link to post
Share on other sites

This is done 5 mins ago on my boot drive. The only difference being ENABLE AHCI in BIOS. There has been no other modifications to nothing. Just a reboot and change a BIOS entry.

BIOS: AHCI

seagate_NCQ.png

BIOS: Disable

seagate_noNCQ.png

Share this post


Link to post
Share on other sites
Seagate is part of what then? You cannot turn AHCI/RAID on or off per drive in BIOS.

I have 3 BIOS options for the ICH9R: IDE, AHCI and RAID.

I have now tested both AHCI and RAID, they are not different if you're using a single hard drive without creating a RAID. The results are identical.

The Intel Matrix Storage Manager is of course installed. Newest version, too, but it doesn't matter, the behaviour hasn't changed for 3 versions now.

And now, forget about copying / transfering or manually trying stuff. Believe me, if it is working it should, you must have no drop in performance in 2 instances of hddspeed. Once you have that, you can manually test real world scenarios.

There is no drop in total throughput with 2 instances of HD_Speed running. Both instances alternate between 0 B/s and ~110 MB/s, with a total of ~110 MB/s. So you would say: it is working.

But nevertheless, real world tests do not show this behaviour!

Two applications, both reading with full speed (each can read faster than the drive can deliver), reach a total throughput of 20-30 MB/s, as I wrote. Believe me, I really wanted it to work, but it simply does not work. The Intel ICH9R is behaving strangely, and HD_Speed is an unfortunate choice to test with, that's all. HD_Speed suggested results that simply aren't there.

By the way, the "TASK MANAGER > PERFORMANCE > RESOURCE MONITOR > DISK"-thing does only work under Windows Vista. I am using Windows XP, and therefore I am using the corresponding XP tool.

And one thing for you to try: try HD_Speed with 128k instead of 256k. The behaviour is completely different for me: both HD_Speeds run at 7-8 MB/s each. With 512k, the behaviour is alternating between those with 128k and 256k, it's quite random.

This could explain why real world behaviour is different, too.

And: please try copying/reading with two applications yourself. Let's see what happens with real world tests on your system.

[Edit] Yeah, it looks like in your first screenshot with AHCI enabled. [/Edit]

Edited by FAT_Punisher

Share this post


Link to post
Share on other sites

@ Fat punisher

Two applications, both reading with full speed (each can read faster than the drive can deliver), reach a total throughput of 20-30 MB/s, as I wrote [.... ]

HD_Speed suggested results that simply aren't there.

:D :D :D

But you should still test with Vista, because XP have no sata support by default (even if you can install "special" drivers).

Edited by extrabigmehdi

Share this post


Link to post
Share on other sites
There is no drop in total throughput with 2 instances of HD_Speed running. Both instances alternate between 0 B/s and ~110 MB/s, with a total of ~110 MB/s. So you would say: it is working.

With which tool are you verifying this with? Please post a screenshot of 2 hdspeed instances and total throughput from XP tool. Test with a newer OS.

My testing platform is 2008 Server x64. Same thing with Vista too in my friend's computer.

Edited by 6_6_6

Share this post


Link to post
Share on other sites
With which tool are you verifying this with?

I have no idea how to translate its name from German to English.

It's part of XP. It's displayed with the "Microsoft Management Console 3.0", and it's probably called something like "Performance".

Please post a screenshot of 2 hdspeed instances and total throughput from XP tool.

It's identical with your screenshots, please simply believe me. I'm already erasing the Seagate for sending it back (I hope they will accept it).

Test with a newer OS.

No.

If it does not work with XP, it's useless for me. I'm not migrating to Vista.

Besides: the "problem" I described (hanging applications etc.) appears under Vista as well. So I bet it wouldn't make any difference.

6_6_6: please do me a favour (and all of the people who are reading this thread): choose a big file (or perhaps two), and run two applications that can read very fast (like copying, or verifying an image or archive). Please verify that every application is fast enough by first running them seperately, and after that, run them together. Monitor the total throughput of the hard drive and please tell us what you get. HD_Speed is not everything.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now