SATA controller speed, JMB 363 vs. Sil 3132

[dieterbsd at engineer.com]

I ran a few quick&dirty performance tests comparing the JMB 363 and Sil 
3132
SATA controllers.  Both are PCIe-x1 cards costing under $US20.  The same
pair of drives was used with both controllers.  System was otherwise 
idle.
Each test was run at least 3 times and the fastest results were used.

FreeBSD 8.0 amd64
Hitachi 2 TB 7200rpm drives
Sil 3132 siis(4)  ada[0-3]
JMB 363  achi(4)  ada[45]

read from 1 disk:
    dd if=/dev/ada5 bs=1m count=1000 of=/dev/null
     jmb363:  128414841 bytes/sec                        7.2% faster 
than sil 3132
    sil3132: 119762153

read from both disks:
     dd if=/dev/ada4 bs=1m count=1000 of=/dev/null & dd if=/dev/ada5 
bs=1m count=1000 of=/dev/null
     jmb363:  193589185 bytes/sec (total)                25.1% faster 
than sil 3132
    sil3132: 154734047

write to 1 disk:
    dd of=/dev/ada4 bs=1m count=1000 if=/dev/zero
     jmb363:  122418113 bytes/sec                        20.1% faster 
than sil 3132
    sil3132: 101934754

write to both disks:
     dd of=/dev/ada4 bs=1m count=1000 if=/dev/zero & dd of=/dev/ada5 
bs=1m count=1000 if=/dev/zero
     jmb363:  175237444 bytes/sec (total)                10.1% faster 
than sil 3132
    sil3132: 159158014

read from 1 disk  &  write to 1 disk:
     dd if=/dev/ada4 bs=1m count=1000 of=/dev/null & dd of=/dev/ada5 
bs=1m count=1000 if=/dev/zero
     jmb363:  231763096 bytes/sec (total)                16.7% faster 
than sil 3132
    sil3132: 198604276

In all cases, the jmb363 outperformed the sil3132.

In all cases, using both disks at once results in less than the expected
performance.

The Sil 3132 can write to both drives faster than it can read from both
drives.  I would expect reading to be at least as fast as writing.