IDE RAID – Is There A Benefit?

Pages: 1 2 3 4 5 6

RAID results

I’m not a big fan of HD Tach other than for checking burst speed (it’s the only one I know of that will test Burst speed), but here I can also use it to confirm (as a second source) my other results. HD Tach 2.61 seems to have just a bit lower maximum Burst score range, so I used the earlier 2.60.

Single drive

2 Drives – Raid 0, Master and Master 64K Block

2 Drives – RAID 0, Master and Slave 64K Block

HD Tach 2.60

Random Access (ms)

10.90

10.10

10.00

Read Burst Sped (mbps)

84.60

90+

90+

Read Speed (kbps)

Max

33103.00

43644.00

43958.00

Min

24319.00

24951.00

24524.00

Avg

25982.00

30111.00

30814.00

CPU Utilization (%)

6.00

11.50

9.40

As you can see, even the use of the earlier version of HD Tach didn’t help. The Burst scores were off the scale, which is a shame because I’d be curious to know what the RAID 0 actually are. Interesting to see the Min. Read speed stayed about the same, Avg. and Max. do show a gain. CPU Utilization did go up a bit with RAID. It doesn’t seem to matter if the drives are on two different ports as masters or master and slave on the same port, except the CPU Utilization being slightly lower if the same port is used (I expected the opposite, especially after looking at the results using a non-RAID setup).

The question is, will Winbench show the same trend?

Single drive

2 Drives – Raid 0, Master and Master 64K Block

2 Drives – RAID 0, Master and Slave 64K Block

Winbench 99

Bus Disk Winmark (kbps)

8260.00

10600.00

7990.00

HE Disk Winmark (kbps)

19300.00

28200.00

21800.00

Disk Transfer rate (kbps)

Beginning

33300.00

59100.00

48300.00

End

33400.00

57300.00

48800.00

Disk Access time (ms)

11.00

10.10

10.60

Disk CPU Utiliz. (%)

3.06

3.61

3.45

Pretty much doesn’t! Clearly shows the Master / Master setup with both drives on different ports as giving the best performance. More of what I would expect. It is Interesting to note that the Disk Transfer rate is almost double when using RAID than a single drive, just what we would hope for and expect if RAID does what it should. Also note that the highest CPU utilization is with RAID and the two drives as master / master on different ports.

But that almost double data transfer doesn’t equate to almost double the disk Winbench. Why? It’s because the disk transfer test is a synthetic benchmark looking for just raw transfer rates and the Winbench score is achieved by using actual data in the way a program (at least those profiled by Winbench) would use it. Good example of ‘Real World’ vs. ‘benchworld’ testing.

OK, now how does that reflect in Winstone? We’ve already seen with the Winbench scores above that ‘Real World’ tests don’t always reflect the scores of a specific system function.

Single drive

2 Drives – Raid 0, Master and Master 64K Block

2 Drives – RAID 0, Master and Slave 64K Block

Bus Winstone 2001

24.70

25.60

25.60

CC Winstone 2001

24.40

25.50

25.30

There is a difference here between using RAID or not, but it’s so small you’d never see it. No difference at all between using two drives on the same port or not. Does that mean RAID 0 offers no benefit? No gain in speed? Only if your use is profiled by the actual test Winstone runs. I would say you need to look at your system usage to determine how RAID 0 would affect your own performance.

So what happens if we change the block size from the default (and max) of 64K to 16K?

2 Drives – Raid 0, Master and Master 64K Block

2 Drives – Raid 0, Master and Master 16K Block

HD Tach 2.60

Random Access (ms)

10.10

10.40

Read Burst Sped (mbps)

90+

90+

Read Speed (kbps)

Max

43644.00

32763.00

Min

24951.00

27580.00

Avg

30111.00

31718.00

CPU Utilization (%)

11.50

11.80

Winbench 99

Bus Disk Winmark (kbps)

10600.00

11700.00

HE Disk Winmark (kbps)

28200.00

29700.00

Disk Transfer rate (kbps)

Beginning

59100.00

58400.00

End

57300.00

60500.00

Disk Access time (ms)

10.10

10.30

Disk CPU Utiliz. (%)

3.61

4.53

Bus Winstone 2001

25.60

25.50

CC Winstone 2001

25.50

25.60

Looks like 16K (smaller) blocks wins. The HD Tach Max. Read Speed with 64K is higher then the 16K score and its result is supported by the Winbench Beginning Disk Transfer rate, but not the HD Tach Avg. or the Min. and the Winbench End score. Just about all the Winbench scores favor the smaller block size, but again the Winstone scores just don’t show any difference.


Pages: « Prev   1 2 3 4 5 6   Next »

Be the first to discuss this article!