Ubuntu Performance Question

Lennart Sorensen lsorense-1wCw9BSqJbv44Nm34jS7GywD8/FfD2ys at public.gmane.org
Wed Jul 11 17:35:26 UTC 2007


On Wed, Jul 11, 2007 at 09:20:13AM -0400, lada-h8kxHjy+vg4AvxtiuMwx3w at public.gmane.org wrote:
> There are two SCSI drives in the P4 box. Here are timing results: 
> 
> lsvatos at grinder:~$ sudo hdparm -t /dev/sda
> 
> /dev/sda:
>  Timing buffered disk reads:  120 MB in  3.08 seconds =  38.96 MB/sec
> lsvatos at grinder:~$ sudo hdparm -T /dev/sda
> 
> /dev/sda:
>  Timing cached reads:   2760 MB in  2.00 seconds = 1379.91 MB/sec

Speed looks fine.  DMA must be on.  And they are not scsi disks, they
just use the libata interface which looks just like scsi, usb, firewire,
and most other disk devices on linux.

Not sure why the P4 would seem slower or under higher load.

Of course if the code you run is very branch heavy, in a very bad case
it is possible the 800MHz P3 could be faster than your 3GHz P4.  I know
a 2GHz Athlon 64 is 5 times faster than a 2.8GHz P4 at running bzip2
compression.  I suspect an 800MHz P3 might actually be close to the
speed of a 3GHz P4 running bzip2.  Long pipelines are that bad at some
types of code since compression is by definition not predicatable to the
branch predictor continuously gets it wrong.  If it was predictable,
then it should be possible to compress it more, hence it must not be
predictable.

--
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists





More information about the Legacy mailing list