Testbed and Methods
To test the actual performance of this external hard disk drive solution we used the following benchmarks:
- WinBench 99 2.0;
- FC-Test 0.5.3;
- Intel IOMeter 2003.02.15.
Our testbed was configured as follows:
- Albatron PX865PE Pro mainboard;
- Intel Pentium 4 2.4GHz CPU;
- IBM DTLA-307015 15GB HDD;
- Radeon 7000 32 MB graphics card;
- 256MB of memory;
- Microsoft Windows 2000 with Service Pack 4.
We tested the performance of our HDD with the help of USB 2.0 (ICH5) interface available on the mainboard, and FireWire interface provided by PCI adapters based on Agere FW323 and VIA VT6307 chips and working with the operation system drivers. For WinBench tests, the drives were formatted in FAT32 and NTFS as a single partition with a default cluster size. In some cases described below we used 32GB logical partitions also formatted in FAT32 and NTFS with a default cluster.
Performance in Intel IOMeter DataBase Pattern
As usual we would like to start with the results our hard drive showed in DataBase pattern. In order to better understand what they mean, we built three graphs characterizing the HDD behavior in this test most vividly.
The first graph of the three shows the HDD performance under linear workload (that is when the queue=1). The performance difference is minimal as we can see, we could even say that there is almost no difference at all. Both interfaces prove equally fast here. I wouldn’t say that any of the IEEE 1394 controllers is preferable here, too. The higher gets the write requests share, the faster works the HDD, which indicates that it uses very efficient lazy write algorithms.
The second graph you can see, stands for queue depth equal to 16 requests. You can see from the pic that FireWire interface looks more attractive then USB 2.0 up to 90% writes share, when all the three graphs merge together. Both IEEE 1394 controllers perform almost equally fast. USB 2.0 controller doesn’t support concurrent requests, so that the performance of the drive working with USB 2.0 interface didn’t change as the queue depth grew higher. At the same time, IEEE 1394 controllers do have this feature. As a result, the HDD runs considerably faster at first, but as the writes share increases, the advantage gets less and less evident, until all three graphs merge together. Judging by this fact I dare suppose that FireWire controllers know to optimize the read requests order.
As the queue depth increases up to 256, we can see how big the advantage of the FireWire interface over USB 2.0 is. It is especially noticeable when the writes share is still pretty low. The performance difference between the two IEEE 1394 controllers is again negligible. As the writes share grows up, the performance gap between the two interfaces reduces.