Performance - Read
I ran the iozone test on multiple configurations, so you'll need to use the info in Tables 2 and 3 to decipher the abbreviations used in the plots below.
|1G||Gigabit Ethernet Adapter|
|Ubuntu||Ubuntu 6.06 Desktop|
|CC||ClarkConnect 3.2 Home Operating System|
|FreeNAS||FreeNAS .671 used|
|XPP||Windows XP Pro Operating System Used|
|S22||Sempron 2200+ CPU on ECS Motherboard|
|D16||Duron 1.6 GHz Processor on ASRock Motherboard|
|RAID5||LSI Logic Megaraid i4 Card Used|
|NoRAID||A single 300 Gigabyte Maxtor ATA drive Used.|
Table 2: Performance Chart Legend Abbreviations
|Terastation noraid 1G||The Terastation NAS System with Gigabit Ethernet Adapter|
|100M Ethernet||This line shows the theoretical throughput of 100 Megabit Ethernet|
|Ubuntu_1G_RAID5_S22||Ubuntu 6.06 DT on RAID5 Hardware with GB Ethernet 2200+ Sempron|
|FreeNAS_1G_RAID5_S22||FreeNAS on RAID5 Hardware with GB Ethernet and 2200+ Sempron|
|Ubuntu_1G_NoRAID S22||Ubuntu 6.06 DT with GB Ethernet and 2200+ Sempron|
|CC_1G_NoRAID_S22||ClarkConnect 3.2 Home on GB Ethernet and 2200+ Sempron|
|XPP_1G_NoRAID_D16||Windows XP Pro with GB Ethernet on a Duron 1.6 GHz|
|Ubuntu_1G_NoRAID_D16||Ubuntu 6.06 DT with GB Ethernet on a Duron 1.6 GHz|
|CC_1G_NoRAID_D16||ClarkConnect with GB Ethernet on Duron 1.6 GHz|
Table 3: Performance Chart Legend
If you are interested in whether the DIY NAS is as fast as entry-level commercial systems, Figures 16 (for Read performance) and 19 (for Write performance) pool the data from my last review and this one. From the graphs and tables the following seems fair to conclude:
• Eliminating RAID5 hardware on a given computer system speeds up your read performance dramatically. At least with the old LSI MegaRAID i4 card, hardware is not an equalizer.
• Eliminating RAID5 hardware will allow you to double file size one more time before poor read performance sets in. The dark blue line that drops between 131072 and 262144 file sizes (128MB and 256MB) represents a drop in performance for the Ubuntu.
• Windows XP Pro's read performance is not significantly different than Ubuntu 6.06 Desktop's peformance with a partial exception of very large data files (1 GB).
Figure 16: All System Read Performance Comparison (click image to enlarge)
Figure 16 is a lot of information to digest, so I've dropped all but the Ubuntu and TeraStation lines in Figure 17. In this figure we have displayed three conditions under which we tested Ubuntu 6.06 Desktop:
- With a Sempron 2200+ and Hardware RAID (blue line)
- With Sempron 2200+ and without RAID(top red line)
- With Duron 1.6 GHz and without RAID (green/blue dashed line)
Figure 17: Ubuntu Read Performance Comparison (click image to enlarge)
The Megaraid i4 card seemed to help the system perform very well at 128 MB file size and below. But, on larger file sizes, the non-RAID configurations were better. The difference between the Sempron 2200+ and the Duron 1.6 GHz can be seen by looking at the difference between the red and green/blue dashed lines. On the faster CPU, hardware RAID was never faster or as fast as not using RAID. I found this to be surprising. I've always thought (hoped) that the hardware cards for RAID were equalizers for the RAID overhead. Not so, particularly for large (i.e., 500 MB and larger) file sizes.
Figure 18 displays the ClarkConnect read performance using AMD Sempron and Duron processors and shows that neither of the ClarkConnect configurations were as fast as the slowest Ubuntu configuration. But surprisingly, the CPU speed difference was much less pronounced in ClarkConnect (difference in blue and yellow lines in Figure 18) than in Ubuntu (difference in top red and green/blue lines in Figure 17).