Gigabyte GA-8KNXP and Promise SX4000 RAID Controller
 

Having read, and experienced, the lackluster performance of the GigaRAID
with 2 Maxtor 160 GB hard drives in RAID 0, I installed a Promise SX4000
RAID controller with 4 Maxtor 160 GB hard drives in RAID 0 on my Gigabyte
GA-8KNXP motherboard.

First, under Windows XP, it has worked well over the past 7 days, 24 hours a
day. No BSODs at all, but then, I am not overclocking (yet).

I tested RAID 0 and RAID 5 configurations with all types of cluster and
striping sizes, and settled on 32 KB clusters and 32 KB striping. I used HD
Tach, DiskSPeed32 and AIDA. I also read all types of Internet reviews on the
SX4000 in different configurations. I settled on this configuration because
(1) I was looking for overall best performance balance between disk reads
and writes and (2) I backup regularly (so RAID 5 redundancy was less an
issue compared to RAID 0).

In my early testing, I quickly dropped RAID 5 for RAID 0, because of the
difference in write performance. For a RAID configuration of 16 KB striped
and 64 KB clusters, using the Random Write test of AIDA32, the average
throughput for RAID 5 was 10.1 MB/s, vs 52.1 MB/s for RAID 0. Changing to 64
KB striped and 64 KB clusters yielded 10.0 MB/s for RAID 5, but RAID 0
dropped to 38.5 MB/s. CPU utilization was about the same for all RAID
configurations. It should be noted that I am using 256 MB of buffer memory
on the Promise SX4000. 4 Maxtor drives were used in all RAID configurations,
of course, RAID 5 used the 4th drive for parity. In general, read
performance of the RAID 5 was 30% slower than the RAID 0 - again, probably
due to having only 3 hard drives for data verses 4 for RAID 0.

I also did RAID 0 single configuration test with a single drive, to explore
performance against a single drive with the other RAID configurations. In
summary, against the RAID 5, the write speed was the same, but the RAID 5
read speed was doubled.

These results would follow popular thinking about performance between RAID 0
and RAID 5. For those following this thread, I can't emphasize that there is
no fault tolerance in RAID 0 - one drive dies, all data is than loss.
Therefore, religious adherence to backup is a must!

Over the years, I have had considerable experience with SCSI RAID systems at
work. At home, on a limited budget, an IDE based RAID system, such as the
Promise SX4000 and possibly other manufacturer's IDE RAIDs, is a cost
effective approach for RAID performance.

I'll check back here regularly to answer any questions - as best as I can.

"Old Dude" wrote in message
link.net...
Having read, and experienced, the lackluster performance of the GigaRAID
with 2 Maxtor 160 GB hard drives in RAID 0, I installed a Promise SX4000
RAID controller with 4 Maxtor 160 GB hard drives in RAID 0 on my Gigabyte
GA-8KNXP motherboard.

First, under Windows XP, it has worked well over the past 7 days, 24 hours
a
day. No BSODs at all, but then, I am not overclocking (yet).

I tested RAID 0 and RAID 5 configurations with all types of cluster and
striping sizes, and settled on 32 KB clusters and 32 KB striping. I used
HD
Tach, DiskSPeed32 and AIDA. I also read all types of Internet reviews on
the
SX4000 in different configurations. I settled on this configuration
because
(1) I was looking for overall best performance balance between disk reads
and writes and (2) I backup regularly (so RAID 5 redundancy was less an
issue compared to RAID 0).

In my early testing, I quickly dropped RAID 5 for RAID 0, because of the
difference in write performance. For a RAID configuration of 16 KB striped
and 64 KB clusters, using the Random Write test of AIDA32, the average
throughput for RAID 5 was 10.1 MB/s, vs 52.1 MB/s for RAID 0. Changing to
64
KB striped and 64 KB clusters yielded 10.0 MB/s for RAID 5, but RAID 0
dropped to 38.5 MB/s. CPU utilization was about the same for all RAID
configurations. It should be noted that I am using 256 MB of buffer memory
on the Promise SX4000. 4 Maxtor drives were used in all RAID
configurations,
of course, RAID 5 used the 4th drive for parity. In general, read
performance of the RAID 5 was 30% slower than the RAID 0 - again, probably
due to having only 3 hard drives for data verses 4 for RAID 0.

I also did RAID 0 single configuration test with a single drive, to
explore
performance against a single drive with the other RAID configurations. In
summary, against the RAID 5, the write speed was the same, but the RAID 5
read speed was doubled.

These results would follow popular thinking about performance between RAID
0
and RAID 5. For those following this thread, I can't emphasize that there
is
no fault tolerance in RAID 0 - one drive dies, all data is than loss.
Therefore, religious adherence to backup is a must!

Over the years, I have had considerable experience with SCSI RAID systems
at
work. At home, on a limited budget, an IDE based RAID system, such as the
Promise SX4000 and possibly other manufacturer's IDE RAIDs, is a cost
effective approach for RAID performance.

I'll check back here regularly to answer any questions - as best as I can.


Hello OD, thanks for the great post .

If it is convenient, I would like to know the Maxtor part number for the 160
GB drives .

It is not clear to me why Gigaraid has a bad reputation. I have 2 WD ATA100
20GB on RAID 0 . They use a 32k stripe.They read at a pretty solid 70 mB/s
on a 64kB data block size.

BTW, you use the term 'throughput'. What is that?

JPS

Hey Old Dude! Just wondering why you did not use the ICH5R controller.
I can see if you want RAID 5 your method is necessary. I take it
those Maxtors are not SATA.

Old Dude wrote:

Having read, and experienced, the lackluster performance of the GigaRAID
with 2 Maxtor 160 GB hard drives in RAID 0, I installed a Promise SX4000
RAID controller with 4 Maxtor 160 GB hard drives in RAID 0 on my Gigabyte
GA-8KNXP motherboard.

First, under Windows XP, it has worked well over the past 7 days, 24 hours a
day. No BSODs at all, but then, I am not overclocking (yet).

I tested RAID 0 and RAID 5 configurations with all types of cluster and
striping sizes, and settled on 32 KB clusters and 32 KB striping. I used HD
Tach, DiskSPeed32 and AIDA. I also read all types of Internet reviews on the
SX4000 in different configurations. I settled on this configuration because
(1) I was looking for overall best performance balance between disk reads
and writes and (2) I backup regularly (so RAID 5 redundancy was less an
issue compared to RAID 0).

In my early testing, I quickly dropped RAID 5 for RAID 0, because of the
difference in write performance. For a RAID configuration of 16 KB striped
and 64 KB clusters, using the Random Write test of AIDA32, the average
throughput for RAID 5 was 10.1 MB/s, vs 52.1 MB/s for RAID 0. Changing to 64
KB striped and 64 KB clusters yielded 10.0 MB/s for RAID 5, but RAID 0
dropped to 38.5 MB/s. CPU utilization was about the same for all RAID
configurations. It should be noted that I am using 256 MB of buffer memory
on the Promise SX4000. 4 Maxtor drives were used in all RAID configurations,
of course, RAID 5 used the 4th drive for parity. In general, read
performance of the RAID 5 was 30% slower than the RAID 0 - again, probably
due to having only 3 hard drives for data verses 4 for RAID 0.

I also did RAID 0 single configuration test with a single drive, to explore
performance against a single drive with the other RAID configurations. In
summary, against the RAID 5, the write speed was the same, but the RAID 5
read speed was doubled.

These results would follow popular thinking about performance between RAID 0
and RAID 5. For those following this thread, I can't emphasize that there is
no fault tolerance in RAID 0 - one drive dies, all data is than loss.
Therefore, religious adherence to backup is a must!

Over the years, I have had considerable experience with SCSI RAID systems at
work. At home, on a limited budget, an IDE based RAID system, such as the
Promise SX4000 and possibly other manufacturer's IDE RAIDs, is a cost
effective approach for RAID performance.

I'll check back here regularly to answer any questions - as best as I can.

The Maxtor hard drives I used was the parallel ATA 160 GB 8 MB cache model:
L01P160.

Choosing this particular drive was easy: I got them on sale/rebates
reasonably cheaply one at a time over a period of 4 months. There was no
thought about their eventual use with the SX4000.

I don't have my original test results for RAID 0 (2 Maxtor 160 GBs) on the
GigaRAID. I just recall that their performance was not as high as I
expected. This may have been unique to me or my PC configuration. As always,
your mileage my vary...

I use the term "throughput" and "transfer rate" interchangeably.

"jpsga" wrote in message
news:sUcsb.127469$ao4.392427@attbi_s51...

"Old Dude" wrote in message
link.net...
Having read, and experienced, the lackluster performance of the GigaRAID
with 2 Maxtor 160 GB hard drives in RAID 0, I installed a Promise SX4000
RAID controller with 4 Maxtor 160 GB hard drives in RAID 0 on my
Gigabyte
GA-8KNXP motherboard.

First, under Windows XP, it has worked well over the past 7 days, 24
hours
a
day. No BSODs at all, but then, I am not overclocking (yet).

I tested RAID 0 and RAID 5 configurations with all types of cluster and
striping sizes, and settled on 32 KB clusters and 32 KB striping. I used
HD
Tach, DiskSPeed32 and AIDA. I also read all types of Internet reviews on
the
SX4000 in different configurations. I settled on this configuration
because
(1) I was looking for overall best performance balance between disk
reads
and writes and (2) I backup regularly (so RAID 5 redundancy was less an
issue compared to RAID 0).

In my early testing, I quickly dropped RAID 5 for RAID 0, because of the
difference in write performance. For a RAID configuration of 16 KB
striped
and 64 KB clusters, using the Random Write test of AIDA32, the average
throughput for RAID 5 was 10.1 MB/s, vs 52.1 MB/s for RAID 0. Changing
to
64
KB striped and 64 KB clusters yielded 10.0 MB/s for RAID 5, but RAID 0
dropped to 38.5 MB/s. CPU utilization was about the same for all RAID
configurations. It should be noted that I am using 256 MB of buffer
memory
on the Promise SX4000. 4 Maxtor drives were used in all RAID
configurations,
of course, RAID 5 used the 4th drive for parity. In general, read
performance of the RAID 5 was 30% slower than the RAID 0 - again,
probably
due to having only 3 hard drives for data verses 4 for RAID 0.

I also did RAID 0 single configuration test with a single drive, to
explore
performance against a single drive with the other RAID configurations.
In
summary, against the RAID 5, the write speed was the same, but the RAID
5
read speed was doubled.

These results would follow popular thinking about performance between
RAID
0
and RAID 5. For those following this thread, I can't emphasize that
there
is
no fault tolerance in RAID 0 - one drive dies, all data is than loss.
Therefore, religious adherence to backup is a must!

Over the years, I have had considerable experience with SCSI RAID
systems
at
work. At home, on a limited budget, an IDE based RAID system, such as
the
Promise SX4000 and possibly other manufacturer's IDE RAIDs, is a cost
effective approach for RAID performance.

I'll check back here regularly to answer any questions - as best as I
can.


Hello OD, thanks for the great post .

If it is convenient, I would like to know the Maxtor part number for the
160
GB drives .

It is not clear to me why Gigaraid has a bad reputation. I have 2 WD
ATA100
20GB on RAID 0 . They use a 32k stripe.They read at a pretty solid 70
mB/s
on a 64kB data block size.

BTW, you use the term 'throughput'. What is that?

JPS

The Maxtors were PATA - I got them before I had considered going RAID,
either SATA or PATA. Those drives came reasonably cheaply.

"Michael Len" wrote in message
le.rogers.com...
Hey Old Dude! Just wondering why you did not use the ICH5R controller.
I can see if you want RAID 5 your method is necessary. I take it
those Maxtors are not SATA.

Old Dude wrote:

Having read, and experienced, the lackluster performance of the GigaRAID
with 2 Maxtor 160 GB hard drives in RAID 0, I installed a Promise SX4000
RAID controller with 4 Maxtor 160 GB hard drives in RAID 0 on my
Gigabyte
GA-8KNXP motherboard.

First, under Windows XP, it has worked well over the past 7 days, 24
hours a
day. No BSODs at all, but then, I am not overclocking (yet).

I tested RAID 0 and RAID 5 configurations with all types of cluster and
striping sizes, and settled on 32 KB clusters and 32 KB striping. I used
HD
Tach, DiskSPeed32 and AIDA. I also read all types of Internet reviews on
the
SX4000 in different configurations. I settled on this configuration
because
(1) I was looking for overall best performance balance between disk
reads
and writes and (2) I backup regularly (so RAID 5 redundancy was less an
issue compared to RAID 0).

In my early testing, I quickly dropped RAID 5 for RAID 0, because of the
difference in write performance. For a RAID configuration of 16 KB
striped
and 64 KB clusters, using the Random Write test of AIDA32, the average
throughput for RAID 5 was 10.1 MB/s, vs 52.1 MB/s for RAID 0. Changing
to 64
KB striped and 64 KB clusters yielded 10.0 MB/s for RAID 5, but RAID 0
dropped to 38.5 MB/s. CPU utilization was about the same for all RAID
configurations. It should be noted that I am using 256 MB of buffer
memory
on the Promise SX4000. 4 Maxtor drives were used in all RAID
configurations,
of course, RAID 5 used the 4th drive for parity. In general, read
performance of the RAID 5 was 30% slower than the RAID 0 - again,
probably
due to having only 3 hard drives for data verses 4 for RAID 0.

I also did RAID 0 single configuration test with a single drive, to
explore
performance against a single drive with the other RAID configurations.
In
summary, against the RAID 5, the write speed was the same, but the RAID
5
read speed was doubled.

These results would follow popular thinking about performance between
RAID 0
and RAID 5. For those following this thread, I can't emphasize that
there is
no fault tolerance in RAID 0 - one drive dies, all data is than loss.
Therefore, religious adherence to backup is a must!

Over the years, I have had considerable experience with SCSI RAID
systems at
work. At home, on a limited budget, an IDE based RAID system, such as
the
Promise SX4000 and possibly other manufacturer's IDE RAIDs, is a cost
effective approach for RAID performance.

I'll check back here regularly to answer any questions - as best as I
can.