Comments on: What nata means for you https://www.dragonflydigest.com/2006/12/05/what-nata-means-for-you/ A running description of activity related to DragonFly BSD. Thu, 07 Dec 2006 19:52:59 +0000 hourly 1 https://wordpress.org/?v=7.0 By: Joe "Floid" Kanowitz https://www.dragonflydigest.com/2006/12/05/what-nata-means-for-you/comment-page-1/#comment-3638 Thu, 07 Dec 2006 19:52:59 +0000 http://www.shiningsilence.com/dbsdlog/index.php/2006/12/05/2006.html#comment-3638 Another question from me out in the peanut gallery…
Has the ATA code grown anything like the “disk-level transaction clustering” patch being experimented with on NetBSD a few years ago?

http://kerneltrap.org/node/410
http://www.blasted-heath.com/nbsd/cluster/

I questioned that at the time (as one of the anonymous cowards in the Kerneltrap thread, no less), but that was before I discovered the areal density of a cheap 120+GB ATA drive with no support for command queuing absolutely stomps any possible benefit from TCQ with older drives on a ‘plain old’ UWSCSI bus. On a single-user desktop, anyway.

On top of that, I now I realize I may’ve misunderstood the thing and it may only be clustering transactions that were already contiguous, but I’m not doing a good job of actually reading the code while trying to get out the door for work. :}

[I do have time to notice that Storagereview.com’s “IOMeter File Server – 64 I/O in IO/Sec” test actually shows a difference between TCQ/NCQ and no queueing, though I also can’t tell if the scores are in any particular real-life units.]

Now watch as I quite probably post this twice. :P

]]> By: Joe "Floid" Kanowitz https://www.dragonflydigest.com/2006/12/05/what-nata-means-for-you/comment-page-1/#comment-3637 Thu, 07 Dec 2006 19:51:24 +0000 http://www.shiningsilence.com/dbsdlog/index.php/2006/12/05/2006.html#comment-3637 Another question from me out in the peanut gallery…
Has the ATA code grown anything like the “disk-level transaction clustering” patch being experimented with on NetBSD a few years ago?

http://kerneltrap.org/node/410
http://www.blasted-heath.com/nbsd/cluster/

I questioned that at the time (as one of the anonymous cowards in the Kerneltrap thread, no less), but that was before I discovered the areal density of a cheap 120+GB ATA drive with no support for command queuing absolutely stomps any possible benefit from TCQ with older drives on a ‘plain old’ UWSCSI bus. On a single-user desktop, anyway.

On top of that, I now I realize I may’ve misunderstood the thing and it may only be clustering transactions that were already contiguous, but I’m not doing a good job of actually reading the code while trying to get out the door for work. :}

[I do have time to notice that Storagereview.com’s “IOMeter File Server – 64 I/O in IO/Sec” test actually shows a difference between TCQ/NCQ and no queueing, though I also can’t tell if the scores are in any particular real-life units.]

]]>