NCQ

[Jimshady]

Just how significant is this?

Let me just rehash what I think it does. NCQ can queue instructions for the next few reads/writes, and can make an informed decision about which is the smarter instruction to fulfill first, based on the rotational position of the heads. Thats all good and well in a synthesised test where the drive is fed a crafted set of instructions that make complete use of NCQ and would die on a non-NCQ drive.

Under what circumstances (in the real world) will NCQ provide real, and significant increases in performance?

I've heard claims that an 7200RPM NCQ disk will perform like a "virtual 10000RPM disk". Meaning that the performance benefit of a 7200rpm physical disk with NCQ will be brought up to par with say a raptor.

I guess the performance benefit would be in when sustained transfers are being interrupted by random reads or something. I don't know. We are going to see more SMP systems when dual-core CPUs are released, so maybe we will be more inclined to do more at once.

Have I missed the point of NCQ? When I first heard about it, I thought WOW, but now I'm a little more skeptical. ¬_¬

 

[Mercutio]

It's something that's of high value to network file/db servers and other devices with an expected high disk load and multiple users.

I wouldn't expect it to be that useful on a single-user desktop PC.

 

[sechs]

Didn't some review show that it may *hurt* desktop performance?

 

[Jimshady]

Good point Mercutio.

 

[Tea]

Errrr .... So what you are saying, Jim, is that NCQ is essentially the old traditional SCSI elevator seeking logic with a dose of steroids? Yes? Or have I got the wrong end of the stick here?

 

[Jimshady]

Yeah you have. Don't worry, Mercutio hit the nail on the head.

 

[Will Rickards]

review of seagate NCQ enabled drive with NCQ supporting controller:
http://www.ocworkbench.com/2004/articles/nativecommandqueue/p1.htm

Interesting increase in file copy results at the end.
Not sure how much faith to put in the results though.
I hope SR reviews it soon.
SR Thread

 

[Jimshady]

Nice. Thanks.


Hey don't OS's provide disk IO queues anyway?

 

[freeborn]

Nice. Thanks.


Hey don't OS's provide disk IO queues anyway?

They do, the hope of NCQ is the assumption that the drive itself knows its physical layout, latencies, and point to point access times and can hopefully make intelligent choices of the order in which to service commands. The OS side queues don't have that insight and must resort to a one size fits all algorythm.

Free

 

[Jimshady]

:O

Ok, because I've been reading a relitively good book called "Operating systems" (duh), and last night I walked away with the impression that the OS had some idea of the rotational position of the heads and actually made a somewhat informed decision about what instruction to send next. Are you just saying that the hardware CQ is just much better and just does a more refined job of what the OS has tried to do?

 

[freeborn]

:O

Ok, because I've been reading a relitively good book called "Operating systems" (duh), and last night I walked away with the impression that the OS had some idea of the rotational position of the heads and actually made a somewhat informed decision about what instruction to send next. Are you just saying that the hardware CQ is just much better and just does a more refined job of what the OS has tried to do?

Actually, I said that is the hope of NCQ. Assuming NCQ is well implemented than yes the drives queueing should be better. Now, how much better? I do not know. The OS side reordering will continue so perhaps the two will compliment each other, perhaps not.

Free

 

[Will Rickards]

Xbit labs tests the Raptor's TCQ capability.
http://www.xbitlabs.com/articles/storage/display/wd-raptor2.html
Read it but the magic number is possible 30% increase.

 

[Jimshady]

The article doesn't lie, but it does fail to address that fact that that simulation is very favorable to TCQ and doesn't at all represent how the disk is actually going to be used.