Time for some more serious benchmarks. Application benchmarks are more realistic than synthetic benchmarks. They try to mimic the behavior of real applications like Photoshop, but also when copying files or swapping. They do this by 'recording' the I/O produced by the actual applications and making a 'trace' of this I/O access pattern. This trace can later be 'replayed' on various hard drives or RAID arrays and see how they perform. This should provide more detailed knowledge of how a drive or array performs on a specific type of I/O but is not always reliable.

The problem of this type of benchmarking is that it implies the application is 100% bottlenecked by the I/O; something that is not always the case. In cases where I/O is only part of the bottleneck, replaying can produce artificially high results on fast storage systems. Fortunately, we are testing a 'bare' disk here so the results should be pretty realistic.


h2benchw
Let's start with a benchmark of German origin - produced by the well-known c't magazin. We will compare the application profile results. Each score represents a certain type of I/O.

h2benchw benchmark (score)
Copying	AAE	108,9
	AAK	114,8
Word	AAE	37,0
	AAK	34,6
Installing	AAE	27,6
	AAK	28,9
Photoshop	AAE	23,9
	AAK	22,8
Swapping	AAE	13,1
	AAK	13,7
F-Prot	AAE	12,4
	AAK	12,9
Benchmark interpretation: higher is better

Interpretation: fortunately for the AAK disk it manages to catch-up and win 4 out of 6 benchmarks.


PCmark
Another popular application benchmark is PCmark. Produced by Futuremark Corporation - the same company that produce the very popular 3DMark benchmark suite for videocards. Again, they test various types of I/O and give a score for each profile.

PCmark benchmark (MB/s)
Virus Scan	AAE	82.5
	AAK	83.6
File Write	AAE	70.9
	AAK	66.2
XP Startup	AAE	10.1
	AAK	11.6
Application loading	AAE	7.6
	AAK	8.3
General Usage	AAE	5.9
	AAK	6.6
Benchmark interpretation: higher is better

Interpretation: again the AAK disk feels at home and wins 4 out of 5 benchmarks.


IOmeter
We will conclude with a semi-synthetic benchmark called IOmeter. Initially crafted by Intel, it is now under an Open Source license and gained wide recognition. IOmeter allows for many adjustments and it can be used to test a very wide range of I/O access patterns. This also has a downside since the added complexity means the application is not easy to use by regular users. Today, we will use it to look at some server-related access patterns.

IOmeter benchmark (I/O per second)
Fileserver	AAE	371
	AAK	396
Webserver	AAE	350
	AAK	370
Workstation	AAE	126
	AAK	125
Random read	AAE	356
	AAK	377
Benchmark interpretation: higher is better
Interpretation warning: results are semi-synthetic in nature and may not tally with real-world performance

Interpretation: interesting results here. The AAK manages to win the server oriented benchmarks but the workstation benchmark ends with a stalemate. The AAK also appears to handle random I/O better than the AAE. This raises questions. Maybe the AAK-disk is optimized for server-type I/O instead?