Performance

We will start our test session with the dependence of the linear read and write speeds on the data block size.

Intel IOMeter: SequentialRead and SequentialWrite

Just in case I would like to remind you that to measure the linear read and write speeds, we use a special pattern for IOMeter utility to send a set of read and write requests to the HDD. Note that the request queue depth is fixed and equals 4 outgoing requests. Once every 30 seconds we change the data block size. This way we get the dependence of the read or write speed on the data block size.

The results obtained with the help of the above described methodology are summed up in the table below (the data are in MB/sec):

Let’s compare the performance of three interfaces with the help of the diagrams below:

As you can see, the situation hardly became any different compared with what we saw last year (see our article called USB 2.0 against FireWire). ATA/100 interface is still beyond any competition, and USB 2.0 still yields to FireWire.

Now let’s check how efficient the controller drivers can cope with the read and write requests for random data blocks. Of course, the results for the DataBase pattern are not evident. You shouldn’t think however, that I am trying to push you to use USB 2.0 and FireWire drives as a solution for DataBase servers. I just want to see, how scalable the interface is depending on the requests queue depth and how greatly the requests processing speed depends on the writes share.