> > > and if you want to give it a try. As to my experience, SCSI stuff is
> > > usually smoother, particularly in respect to CPU load.
> >
> > please don't repeat this trite old falshood. modern ATA controllers
> > are just as busmastering as decent SCSI cards.
>
> So please name:
> - ATA implementations that do tagged command queueing
none. but TCQ has NOTHING to do with busmastering. under Linux,
and all other nontrivial OSs, the kernel does head scheduling for IDE.
this works extremely well, since linear address is monotonic with
head position (ignoring bad-block remapping).
> - ATA implementations that do dis-/reconnect in the same channel
current IBM disks support this, and there's no controller support necessary.
our driver doesn't support it yet either. again, it has nothing to do
with busmastering, and indeed since you get 2 or 4 free ATA channels,
and can add them for about $10 apiece, this is almost irrelevant.
> To do busmastering, you also need a drive that actually offers this in
> a decent manner.
no, busmastering is a property of the controller. just as there exist
plenty of non-busmastering SCSI controllers.
> The Pioneer DR-A14S does not offer UDMA, just to name one.
does it do any DMA mode? UDMA is much better, because it CRC's everything,
but 16MB/s busmastering mode2 DMA is widely available and plenty fast for
cdroms.
> I'll see if I can get hold of decent bonnie(++) figures, ripping my
> systems apart and reconfiguring in case this is of interest, I'll
> however not do so unless asked to.
don't bother. here's a recent but low-end Maxtor 91021U2 doing
UDMA33 on 2.3.99-pre6-something:
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
500 15233 94.6 17679 19.8 5378 8.6 9288 43.2 23936 22.2 105.5 1.2
(that was on a 128M machine using BX's PIIX4 controller). note that the %CPU
scores reported by Bonnie are just plain wrong. I've measured the actual
overhead around 5%, computed from the observed performance cost to a
CPU-bound benchmark.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Apr 23 2000 - 21:00:15 EST