4. Performance - Write

I used IOzone to give the Office a performance workout. The full testing setup and methodology are described on this page, so I'll just concentrate on the results here. I should note that, thanks to IOzone's current maintainer Don Capps, I'm somewhat savvier than I was for my ioGEAR BOSS vs. Tritton ASAP review about how to use IOzone properly!

Figure 4 shows a 3D surface plot of the Office's NDAS (Ethernet-connected) write performance using file sizes from 64 kByte to 1 GByte and record sizes from 4 kByte to 16 MByte. Note that the plot is oriented so that larger file sizes are closer to the front.

Figure 4: Write performance - NDAS mode
(click on the image for a larger view)

This time I know that the "mountain" rising out of the plot's plateau is due to cacheing in the Win XP test system and Office's hard drive 8 MB buffer. You can see that small record sizes hold back performance, with peak 496,000 kBytes/sec speed obtained using a 128 kByte record size and 8 MByte file size.

NOTE! The flat spots at the left front and right rear of the plot are zero value areas due to the IOzone test parameters that I set and not due to performance problems.

Since the cached performance expands the scale such that you can't see the hardware-limited performance details, I've generated a second plot that zeroes in on that area.

Figure 5: Write performance detail - NDAS mode
(click on the image for a larger view)

Figure 5 shows Write Transfer Speed vs. Record size for file sizes from 32 MByte to 1 GByte. IOzone's rule of thumb for seeing non-cached performance is to look at results for file sizes greater than the amount of RAM in the system running IOzone (504 MB in this case). But Figure 5 shows non-cached performance kicking in even with a 32 MByte file size. Even though the non-cached performance comes in around 7700 kBytes/sec, that's not bad given that the max raw data rate for 100Mbps Ethernet is 12500 kBytes/sec.