Re: 8 milliseconds minimum gap between two consecutive disk I/O

Andre Hedrick (andre@suse.com)
Mon, 3 Jan 2000 14:40:28 -0800 (PST)


On Mon, 3 Jan 2000, Richard B. Johnson wrote:

> On Mon, 3 Jan 2000, Azeem Shahjahan Jiva wrote:
>
> > +---------- Thus spoke Juan Piernas Cánovas on Mon, Jan 03, 2000
> > |Also, I've drawn from the kernel source that Linux isn't able to deal
> > |with two simultaneous disk requests, each on a separate disk, if both
> > |disks are on the same IDE ribbon (that is, they use the same IDE bus).
> > |Am I wrong?.

hdparm -t /dev/hda ; hdparm -t /dev/hdb

/dev/hda:
Timing buffered disk reads: 64 MB in 6.78 seconds = 9.44 MB/sec
/dev/hdb:
Timing buffered disk reads: 64 MB in 7.17 seconds = 8.93 MB/sec

hdparm -t /dev/hda & hdparm -t /dev/hdb &

/dev/hda:

/dev/hdb:
Timing buffered disk reads: 64 MB in 14.39 seconds = 4.45 MB/sec
Timing buffered disk reads: 64 MB in 14.64 seconds = 4.37 MB/sec

You get about a 50% drop in throughput.

HDA & HDB are first generation ATA-33

./hdparm -t /dev/hda ; ./hdparm -t /dev/hdc

/dev/hda:
Timing buffered disk reads: 64 MB in 4.75 seconds =13.47 MB/sec
/dev/hdc:
Timing buffered disk reads: 64 MB in 4.91 seconds =13.03 MB/sec

./hdparm -t /dev/hda & ./hdparm -t /dev/hdc &

Timing buffered disk reads: 64 MB in 7.60 seconds = 8.42 MB/sec
Timing buffered disk reads: 64 MB in 7.79 seconds = 8.22 MB/sec

HDA & HDC are third generation ATA-33

> > Hi,
> > I was under the impression that was an IDE limitation and not an OS one.
> >
>
> IDE doesn't provide for multiple queued commands, disconnect, reconnect,
> etc. If you want this, you use SCSI. IDE is like a big floppy.

That is because I have not had the time to add the "multiple queued commands"
updates. Second, drives are just now comming to market to support this
fine feature. Just so you know, I have the luxury of merry gang at
QUANTUM Drives to run EIDE analize inline probe to ensure that I get it
correct. Since all of there latest 7200RPM EIDE disk line are SCSI cores,
platters, heads, spindle, etc are identical, the quality issue is now a
moot point.

Andre Hedrick
The Linux IDE guy

-
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/