QoS Performance Tests - Priority Features
Since the GS5 can be used as a standalone switch and because I didn't want the slower wireless link to interfere with testing, I ran Ethernet-only tests of its QoS features.
One of the things that I found is that you can't set the QoS features so that the GS5's switch behaves completely like a normal switch. I set up a three computer test configuration:
- Computer 1 - Athlon 64 3000+, 512MB, Win XP Pro SP1
- Computer 2 - Pentium 4 2.4GHz, 504MB, Win XP Home SP1
- Computer 3 - Celeron 1GHz, 576MB, Win XP Home SP1
and connected them to Ports 1, 2 and 3 of the GS5. I reset the GS5 to defaults, which set all the occupied ports to High priority, using the Weighted Fair Queue method. I then set up an IxChariot test that ran simultaneous TCP throughput scripts from the two slower to the one fastest machine. The only change I made to the throughput script was to raise the transfer file size from 100,000 to 1,000,000 Bytes.
Figure 10 shows that the two computers did not get equal shares of bandwidth, even though port priorities were set equally. I also tried setting all ports from High to Critical, with similar results. The exact throughput graph changed from run to run, but in no case did I get the expected equal-sharing of throughput. (Note that there is no way to completely disable the GS5's QoS settings.)
Figure 10: Two-to-one throughput test - GS5
(click on the image for a larger view)
To ensure that there wasn't something in my test configuration that was causing the observed behavior, I kept everything the same except for moving all three machines to three ports on my normal 10/100 LAN switch. Figure 11 shows the results of the move, which is the exact behavior I'd expect from a switch, i.e. equal sharing of bandwidth.
Figure 11: Two-to-one throughput test - regular 10/100 switch
(click on the image for a larger view)