Re: Can't disable NCQ, /sys/block/sdc/device/queue_depth is readonly
On Sat, May 05, 2007 at 06:58:29PM -0700, Roy Franz wrote:
>
> 1) After coming across a simple random seek benchmark, I am finding that my
> WD raptor 10K drive is giving me about 10 ms seek time, where other raptors
> benchmarked with the same program are reporting 7.5 - 8.0 ms. This is
> a fairly substantial difference, and I am trying to track it down. (As
> an aside, I found that
> enabling acoustic management made a great difference on several drives
> - about a 30% increase is seek time.)
>
> 2) From my reading about NCQ, one of the quirks of Raptors is that
> when NCQ is enabled, the drive disables its internal readahead. This
> results in significantly reduced sequential throughput. Apparently
> the windows drivers have some workarounds to deal with this.
>
>
> I'l like to be able to test to see what effect NCQ has on these
> issues. I find it very strange that the official interface for
> controlling this seems to be broken (or misconfigured on my system),
> and I could not find anything useful about this
> via google. (I did find a couple other posts about the same problem,
> but no resolution.)
So are you wanting a fast responsive multuser, multitasking, computer
system or are you wanting a single disk to perform well on a single
benchmark?
NCQ is about enabling an ATA disk to handle requests for different
segments of the disk. The test in 1. probably doesn't issue a bunch of
seek requests at once and average the time, rather it probably
sequentially issues seek requests and times the results then averages.
If you have an IO bottleneck on your computer, try to determine its
nature. If it occurs during a pure sequential read (with no other
competing drive requests from other processes) then look at the data
transfer rate and compare that with the drive spec, and the MB's spec.
As for accoustical measures, I believe that they depower the seek motor
which makes the head arm move slower, lengthening seek time.
Doug.
Reply to: