Saber Cherry wrote:
I suppose you could try to copy two large (~1GB+) directories at the same time (two file copy windows), and see if it takes far longer than it would take to do them one after another. With NCQ working it should take equal or possibly less time to do them in parallel than in series. Without NCQ, it should take equal or more time in parallel. That makes it hard to test for certain, but on the plus side, it should only take a minute (and I really appreciate your offer). As for XP, that's fine - maybe the board works perfectly in XP but not in Vista. Since I will never buy Vista nor tell anyone else to, I don't really care
What I really want to know is whether, when I am doing disk I/O (say, zipping a huge folder), and I start doing additional disk I/O (launching an application), all of a sudden both tasks will take forever to complete, like on my current computer. Because I REALLY hate that. My main HDD is ATA-66; pre-NCQ.
NCQ is not going to have much impact on either of those tests. There just won't be enough outstanding I/Os for re-ordering to make a difference. It's only on server loads that it matters (and no, "doing a bunch of things at the same time" on a desktop does not approximate a server load in this regard -- simultaneously zipping to a file and launching an app will result in just a handful of concurrent I/Os). Stepping up to any
modern drive from an ATA-66-era HD is going to be such a huge improvement that NCQ will be lost in the noise.
You will need AHCI if you want eSATA (well, if you want to be able to hot-plug/unplug it -- though note not all hardware supports that anyway).
As for NCQ being a wash: almost every performance problem I have experienced on a modern computer, aside from 3D game framerates, has been largely due to naive/simple/stupid/broken HDD control algorithms that make disk I/O throughput go at under 1% of its peak. My usage patterns cause my HDDs to routinely become massively fragmented, so badly that the crippled XP defragmenter cannot help (but fortunately I found quality freeware that works).
Then you misunderstand NCQ. Single-user operations, even when you're multitasking, involve relatively few outstanding I/Os -- and NCQ works by reordering I/Os. Not many I/Os...not much NCQ can do. Even if you have a heavily fragmented disk, and you've got a program doing something disk-intensive, NCQ rarely helps. If the system is trying to read or write a file on the disk and a sector goes by that happens to be from somewhere later in the file, the drive won't pick it up because it doesn't know it is needed yet. NCQ doesn't help with that, because you're still dealing with just a single I/O. On the other hand, if you're running a web server and there are 30 active threads, and each of them happens to be asking for a file, NCQ will re-order those 30 outstanding I/Os so it picks up clusters to service one request "in passing" while it is moving the head to service others. NCQ works when you have many outstanding I/O requests
, not when you have just a handful (even if those happen to be requesting very large or heavily-fragmented files). NCQ can even hurt in the single-user case
where there are few I/Os (though that's less of an issue than it was with older drives in the early days of the tech, you can still see it in the linked storage review tests); on the other hand, if you regularly have more than 30 outstanding I/Os you should be looking at SCSI drives that do it even better.
If your usage pattern leaves your HD heavily fragmented, then (a) you need a larger disk (more free space slows down the fragmentation process, and makes defragmentation easier) and (b) you need to invest in some good defragmentation software -- and that tends not to be free. You might also want to consider a 10+K drive, since fragmentation increases seeks and seek performance is related to spindle speed (and the higher RPM drives have faster heads too, over and above the lowered rotational latency).