Pros and Cons of amd64 (versus i386).
Matthew D. Fuller
fullermd at over-yonder.net
Mon Apr 10 16:18:11 UTC 2006
On Mon, Apr 10, 2006 at 10:02:34AM -0400 I heard the voice of
Christian Lopez de Castilla Wagner, and lo! it spake thus:
>
> As you can see, the outside is more than twice as fast in this case.
> Just a guess, since both are IBM disks: You're using a
> Workstation/Server disk, which probably performs in a more balanced
> way across the platter, while this (consumer disk) is not
> performance-oriented. Maybe SCSI and IDE devices are not as similar
> as we all thought?
I would think it's more a matter that it's just a much older and much
lower-density disk, so the difference isn't as pronounced. The seeks
show where the SCSI pulls ahead of the IDE. For instance:
> Full stroke: 250 iter in 5.650173 sec = 22.601 msec
Full stroke: 250 iter in 3.986904 sec = 15.948 msec
> Half stroke: 250 iter in 4.581880 sec = 18.328 msec
Half stroke: 250 iter in 3.314294 sec = 13.257 msec
> Quarter stroke: 500 iter in 7.510191 sec = 15.020 msec
Quarter stroke: 500 iter in 5.683162 sec = 11.366 msec
Now, compare a modern SCSI drive:
Seek times:
Full stroke: 250 iter in 2.144289 sec = 8.577 msec
Half stroke: 250 iter in 1.649111 sec = 6.596 msec
Quarter stroke: 500 iter in 2.707602 sec = 5.415 msec
Short forward: 400 iter in 1.726583 sec = 4.316 msec
Short backward: 400 iter in 1.076023 sec = 2.690 msec
Seq outer: 2048 iter in 0.198563 sec = 0.097 msec
Seq inner: 2048 iter in 0.194318 sec = 0.095 msec
Transfer rates:
outside: 102400 kbytes in 1.406852 sec = 72787 kbytes/sec
middle: 102400 kbytes in 1.557666 sec = 65739 kbytes/sec
inside: 102400 kbytes in 2.112654 sec = 48470 kbytes/sec
da0: <SEAGATE ST373453LC 0006> Fixed Direct Access SCSI-3 device
da0: 70007MB (143374744 512 byte sectors: 255H 63S/T 8924C)
The difference in transfer rates is less than half there, too. But,
this is a 15k RPM drive, which means that the platters are much
smaller, so there's much less difference in the linear speed between
the rim and the spindle side. It's not because of any attempted
compensation toward an "average" performance, it's just a side effect
of trying to not require a nuclear reactor to power it 8-}
--
Matthew Fuller (MF4839) | fullermd at over-yonder.net
Systems/Network Administrator | http://www.over-yonder.net/~fullermd/
On the Internet, nobody can hear you scream.
More information about the freebsd-stable
mailing list