Zentraedi

Big Difference With Tcq And Raptor

Recommended Posts

HOLY CRAP was there a big difference after enabling TCQ on my Highpoint 1820 with a WD740 Raptor. It had no effect on my IDE RAID with IDE to SATA convertors though.

Also, the Raptor is my boot drive and it took significantly longer to boot.

Raptor without TCQ:

wow.jpg

Raptor with TCQ:

after.gif

Ok, maybe other benchmarks may be in order, but the Raptor definitely feel much slower anyway.

Share this post


Link to post
Share on other sites

Only tasks that produce deep, disparately targetted (in physical space) queue depths will benefit from TCQ.

IOMeter is a benchmark which will produce such an access pattern, and it is free. Run it to see relevant data.

It seems unlikely ATACQ could reduce performance. But I guess it must reduce HDTach performance. That HDTach demonstrates reduced performance is likely only indicative of what a terrible benchmark it is. That you seem to have observed a noticeable decrease in performance is very strange.

A thorough examination of Highpoint's website revealed no suggestion that the controller supports ATACQ. I don't believe that any Highpoint chipsets support ATACQ. I cannot find it mentioned on any of their controllers or chipsets. How did you enable it?

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

I enabled TCQ through a registry key. It required rebooting for changes to take effect though.

Here's what's in the driver's readme file:

5. Driver Parameters

=====================

On Windows NT/2000/XP/2003 system, you can pass parameters to driver by the

following registry key:

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Service\hptmv\Parameters\Device

Under this key, create a string value named "DriverParameter" and set the

value to the parameter string (without quotition mark).

Available parameters:

"TCQx=1" or "TCQx=0"

Enable/disable TCQ support, where x is channel number - 1 (0~7).

"smart=1" or "smart=0"

Enable/disable S.M.A.R.T support.

Multiple settings can be seperated by semicolon, e.g. "tcq1=0;tcq2=1;smart=1".

Share this post


Link to post
Share on other sites
I am just guessing, but maybe the TCQ implementation in the 1820 is flawed?

I'm guessing HDTach is flawed

Well, I noticed boot up with Raptor was much slower as well as everything else too.

Come to think of it these were the same problems I had when I first got this card and was using the older bios and drivers.

Share this post


Link to post
Share on other sites

If you are forcing a protocol that your controller cannot support it may run the bus in a legacy mode. If nothing else commands may not be fulfilled and have to be reissued causing performance issues. If the controller doesn't support the return of commands out of order it may be reissuing requests whenever the drive reorders them; this would certainly cause a significant penalty.

The Highpoint 1820 makes no claim of ATA Command Queuing support and I would bet that this is the core of the problem.

I would set things back to normal and see if things improve. I agree that HDTach is a pretty useless benchmark. It should see no penalty however. TCQ should be transparent to it, and, additionally, probably wouldn't even be used under the loads that it generates. That HDTach oberves such an extreme penalty is a fairly certain suggestion that something is wrong.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites
If you are forcing a protocol that your controller cannot support it may run the bus in a legacy mode.  If nothing else commands may not be fulfilled and have to be reissued causing performance issues.  If the controller doesn't support the return of commands out of order it may be reissuing requests whenever the drive reorders them; this would certainly cause a significant penalty.

The Highpoint 1820 makes no claim of ATA Command Queuing support and I would bet that this is the core of the problem.

Well, the RAID 0 I have with 2 IDE drives (each with RocketHead IDE to SATA adapters) had no change in performance with or without the option enabled. Meanwhile, the Raptor did have issues.

As for TCQ support, the driver README file tells how to enable this and it say it right on the box the controller came in!

sata.jpg

Share this post


Link to post
Share on other sites
and it say it right on the box the controller came in!

That does appear to be the case :)

I would then guess that the HPT driver blows goats .... combined with HDTach blowing sheep.

Share this post


Link to post
Share on other sites
and it say it right on the box the controller came in!

Indeed. I'm surprised they didn't mention it on the website.

By Serial ATA Command Queuing they almost certainly mean ATA Command Queuing. ATA Command Queuing is the Tagged Command Queuing protocol used in later ATA iterations and in SATA 1.0.

Serial ATA 2.0 will feature a new specification, Native Command Queuing, in addition to retaining legacy support for ATACQ.

Under the light of this confirmation I am very surprised by the results. It is likely a bug in HDTach.

I don't know how you could blow TCQ since the drive does all the work. All the controller has to do is support setting its BSY (busy) bit to 0 in response to the drive declaring it has queued a command instead of waiting for the drive to finish completing the work. This allows it to send more commands and allows the drive to reorder them.

I would check performance using other tools. Are you sure there is a noticeable difference in real world performance with it enabled vs having it disabled? I definitely recommend using IOMeter. As well try Winbench99. If you register a performance penalty in either benchmark relative to having TCQ disabled I would certainly report it as a bug to Highpoint's technical support, because there is almost certainly a problem with something they are doing.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

Well, I sent an email to Highpoint and here is the response I got:

Dear Justin,

It is a known problem on WD740GD, but there is no this issue on other TCQ-supported hard drive, such as IBM hard drive.

Please try to contact the WD to update the firmware.

Regards,

HighPoint Technologies Inc.

Please include all emails in all correspondence so that there is a complete record

At 13:02 2004-2-10, you wrote:

>Well, I tried enabling TCQ via the Registry Key as per the instructions in

>the README file of the latest drivers.  To my surprise this lowered the

>performance of the Western Digital WD740GB Raptor by 10 fold!  This is

>basically the way things were before.  The only conclusions I can think of

>are some sort of incompatibility with TCQ and this particular drive or a

>flawed implementation of TCQ on the controller's side.

>

>If you could shed some light on to my situation it would be much

>appreciated.

>

>Thanks

An issue with the Raptor huh? Guess I'll be sending an email off to Western Digital now. Hopefully, this gets fixed soon.

Share this post


Link to post
Share on other sites
Dear Justin,

It is a known problem on WD740GD, but there is no this issue on other TCQ-supported hard drive, such as IBM hard drive.

An issue with the Raptor huh? Guess I'll be sending an email off to Western Digital now. Hopefully, this gets fixed soon.

I googled "ATA command queuing" yesterday. The FreeBSD ATA driver gets the most hits. In Google Groups Highpoint gets a lot. They seem to be only commerical product with support, yet their web site doesn't mention it.

Since there are ATA controllers and IBM disks that support Command Queuing, it would be nice to see some benchmarks. All I found was dome "dd" tests on BSD.

Share this post


Link to post
Share on other sites

Well, hopefully this issue gets resolved soon. I just check the status of my problem on Western Digital's site and they upgrade the Status to "Escalated". I wonder if that means they're working on a firmware or just forwarded this to a higher level tech person?

Share this post


Link to post
Share on other sites
Well, hopefully this issue gets resolved soon.  I just check the status of my problem on Western Digital's site and they upgrade the Status to "Escalated".  I wonder if that means they're working on a firmware or just forwarded this to a higher level tech person?

let us know, maybe you get to flash a new firmware?

Share this post


Link to post
Share on other sites
I just check the status of my problem on Western Digital's site and they upgrade the Status to "Escalated".

Could be good, could be bad. On our IT helpdesk, a call gets "escalated" if you ignore it for too long. At Western Digital, it probably means they've noticed that it's not a problem their regular call centre staff can fix. It's anyone's guess how much attention/techie effort it's now getting.

Share this post


Link to post
Share on other sites

The Vaporware-ish RAIDCore controllers that Tom's readers got excited about for a while also support ATA CQ, but really this isn't an issue for anything but server use. It is interesting that the issue being discussed in this thread did not appear in the Tom's review. No doubt their ineptitude kept them from learning of the feature and turning it on. Of course it may be pretty much benchmark specific like the SCSI issue in XP was with HDTach (or was it ATTO?).

Desktop programs wait to recieve any data access they make before sending the next one. The way they are coded prevents them from generating any sort of queue. The vast majority of us are OK. Only software like a database or webserver who's I/O are generated by multiple users, at the whim of the users, will ever produce a queue that can be reordered.

It's too bad the 'feature' is actually a performance killer, but you'll not lose any performance running the WD740GD with it off. It may slow the Raptor's enterprise server penetration, but I haven't seen any major vendors pulling SCSI from their servers and replacing it with Raptors so that may be a non-issue. Any big vendors who use SATA seem to be using it for the cheap drives, not for the performance. I am positive this won't change until SATA II and SAS.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

I find this really fascinating. It is nice to see some hard evidence on this subject - after our discussion about the SATA specs and if the Raptor command queuing would work.

Is the card based on the Silicon Images 3124? If so, this would be big problem. But, then again, we don't know if there actually *is* a problem until the server benchmarks are run. :)

I also love the comment "known issue." Well, thanks for sharing. I would think Highpoint has more to lose in this scenario since most people would swap drive controllers (e.g. to another brand) before trying other drives.

Even though I think labeling the Raptors "Enterprise-class" was more of a marketing gimmick (like "Pro" or "Ultra"), lack of Native Command Queuing (much less functional ATA Command Queuing) will be a wall if they are really after the server market.

However, I don't think the low-end server market really cares.

[if I was still an Enterprise CIO, I'd be waiting for SAS anyway.]

I still would not lay the blame solely on Western Digital. Over the decades I have run into many drive / controller mismatches. Drives "fail" daily on one brand of controller, and work flawlessly on another. Made me wonder what part of SCSI was standard. :)

You would need one of those "IBM drives" to determine if Highpoint really supports command queuing. At this point it is all hearsay.

Dogeared

8^)

Share this post


Link to post
Share on other sites

While I can't find a good enough photo to tell or specifications on their site, Highpoint has always used their own controllers on their RAID cards in the past. Their primay business is selling their chips after all, so I am fairly certain the SI 3124 is safe.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites
While I can't find a good enough photo to tell or specifications on their site, Highpoint has always used their own controllers on their RAID cards in the past.  Their primay business is selling their chips after all, so I am fairly certain the SI 3124 is safe.

Do well.

Jonathan Guilbault.

Agreed.

Thought I would ask anyway, since I could not find any info on Highpoint silicon during our last thread. :D

Usually manfacturers are interested in publishing new products.

Dogeared

Share this post


Link to post
Share on other sites

The Highpoint 1820 uses a Marvel chip.

http://www.marvell.com/products/storage/sata/88SX5080.jsp

Marvel has always been strong in SATA so I'm surprised that they'd have a flawed implementation of command queueing... parhaps it is the Raptor's firmware that is flawed.

Has anyone seen any reviews, benchmarks, or boards for sale based on the new Silicon Image 3124? Besides this Tom's Hardware review that mentioned its existence, I have not.

BTW, the Marvel is also used on this SuperMicro PCI-X SATA RAID board (the non-bootable version).

Share this post


Link to post
Share on other sites
I could not find any info on Highpoint silicon during our last thread.

Ya, they actually don't appear to have a SATA IC announced at the moment. I believe the 1820 must use bridge chips; they're visible on it's little brother, the 1640.

For the 1820 you can actually tell from the photo on their website that it is highpoint logic on the board (I had to lower my resolution to blow it up). I believe the bridge chips must be on the backside of the 1820, I'm sure Zentraedi could enlighten us.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

Gilbo,

Actually, the HPT uses the Marvel IC (see my prior post), and that is a native SATA chip so no bridge chips are necessary.

I get the feeling Highpoint is taking a diff. direction with this 1820 controller, not relying on its own silicon (or board design even -- compare it to the Supermicro card).

I also read today that LSI will have a controller based on the Marvel in Q3 of this year. That's a vote from the "professional" end of the market.

s.

Share this post


Link to post
Share on other sites

shoek,

Firstly, thanks for the info on the controller. I was wondering how it was working because, as I said, there were no bridge chips visible on the 1820.

Marvel has always been strong in SATA so I'm surprised that they'd have a flawed implementation of command queueing... parhaps it is the Raptor's firmware that is flawed.

From this discussion it seems like Highpoint is laying the blame squarely on Western Digital, and judging from the way Western Digital tech support has been dealing with it it seems that that may be the case.

Do well.

Jonathan Guilbault.

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