Originally
posted 07/18/03 by rob-ART
morgan,
mad scientist
This the
continuation of the article is comparing the new
AEC-6885MS
4-CH High Scalability ATA133 RAID PCI Host
Adapter
from Acard.
We're comparing it running 4 drives in RAID 0
configuration to FireWire 800 four drive RAID 0
(both 2 and 4 channels) and Ultra320 SCSI four
drive RAID 0 (both 2 and 4 channels).
REAL WORLD
TESTS
ANALYSIS
1. Is the
AEC-6885MS
4-CH High Scalability ATA133 RAID PCI Host
Adapter
card faster than two 2 channel AEC-6880M
HW RAID cards? In
4 out of 7 tests, the 6885MS was significantly
faster.
In the remaining tests, it was slightly slower. Add
to that the fact that it uses only one slot and
supports three modes of HW RAID and the 6885MS
becomes a clear winner over its
"brother."
2. Which
produces a faster 4 drive striped array (RAID 0),
ATA-133? FireWire 800? or Ultra320 SCSI?
Surprise. The
FireWire 800 four drive, four channel setup was the
clear winner
in 4 out of 7 tests. And it was a close second or
third in the other three.
3. If cost is
the factor, setup is the best buy? If you have
room inside your tower for four drives in addition
to the boot drive, the Acard 6885MS scenario is the
least expensive. The FireWire 800 scenario costs
more because of the required enclosures, but is
cheaper than Ultra320 SCSI. Plus it has the
advantage of "living" outside the box. Therefore,
I
rate FireWire 800 RAID the best buy.
4. So why
didn't the 320MB/s Ultra SCSI setup blow the others
away? It's
partly the fault of drive
technology.
Though the SCSI drives ran at twice the rotation
speed of the ATA drives (15,000rpm vs 7200rpm) and
require half the average access time (3.6ms vs
8.5ms), that just wasn't enough. They are not twice
as fast. More like 38% faster. (75MB/s vs 54MB/s
sustained READ). And though each channel can handle
three times the throughput (320MB/s vs 100MB/s or
133MB/s), apparently that's not enough either.
I
think the main bottleneck is the Power Mac G4's PCI
bus controller.
That's makes sense when you consider I used the
same UL4D controller on a Dual Xeon with 100Mhz
PCI-X slot and got TWICE the speed.
5. Take another
long look at the graph above to see what the
G5
Power Mac
potentially offers us, thanks to the new 100MHz and
133Mhz PCI-X slots. It will be interesting to see
how well the parallel and serial ATA host adapters
do on the PCI-X bus. We've already tested a
multi-channel FireWire 800 PCI-X card in the 100Mhz
PCI-X slot of the Dual Xeon. But it was no faster
than using multiple single channel cards on the G4
Power Mac.
TEST
NOTES
The Power Mac
G4/1.42MP with 2GB of DDR memory and OS X (10.2.6)
was the test mule.
CONTROLLERS AND
DRIVES TESTED
The
AEC-6885MS
4-CH High Scalability ATA133 RAID
Adapter
connected to four drives (two Hitachi
180GXPs and two Maxtor
DiamondMax Plus 9s). AEC-6885MS can be ordered
direct
from Acard
or dealers listed on their site. Your favorite
Ultra ATA drives can be ordered from
OWC,
TransIntl,
and GoogleGear.
Two
AEC-6880M
dual channel ATA HW RAID Adapters connected to four
drives (two Hitachi
180GXPs and two Maxtor
DiamondMax Plus 9s). The four drive, four channel
stripe was created by taking two HW striped pairs
and making them a SW striped pair using Apple's
Drive Utility. AEC-6885MS can be ordered
direct
from Acard
or their dealers.
The four channel,
four drive FireWire 800 RAID was created using
three FW800 PCI cards in conjunction with built-in
FW800. (Using four PCI cards was much slower than
three with the internal FW800 port.) The four
enclosures included two Mercury Elite Pro FW800
enclosures from Other
World Computing
and two IceCube800 enclosures from FWDepot.
The four drives were Hitachi
180GXPs.
The two channel
FireWire 800 RAID is similar to above except only
two PCI cards were used.
The four channel,
four drive Ultra320 SCSI array was created using
both the ATTO
UL3D and UL4D dual channel Ultra320 controllers
with one drive on each channel. Two drives were
Seagate
Cheetah 15K.3 and two were Maxtor
Atlas 15K.
The two channel
Ultra320 SCSI array was done by putting two drives
on each channel. Results for both the UL4D and UL3D
are posted to show that the UL4D is overkill on the
G4 Power Mac. But on the G5 Power Mac, you should
see it pull away.
SEE
"HOW
WE TEST"
for details on the tests reflected in the
graphs.
|