Sunday, January 2, 2011

The Importance of Benchmarks, Part 2

Happy New Year!

I tried running my Thinkpad (T500 with Intel 5300 wireless card) using 802.11n and found results extremely variable.  At times it would get terrific throughput. At other times it could not hold a connection. Some research indicated that there is a problem related to Intel firmware with the drivers released with Ubuntu 10.10. It's not clear if the same problem exists with 10.04 which I'm currently running. What I do know is that the techniques (described for 10.10) to force the driver to use G only do not work on my Thinkpad. I have to configure my router as G only in order to get a solid wireless experience.

This makes me wonder if there was actually a problem with my older router. I started having connection problems with it and perhaps that was a result of driver/firmware problems and not a malfunctioning router. I should plug it in and test it out one of these days.

Part 2b

I realized that my new WiFi router (D-Link DGL 4500) supports gigabit Ethernet. So do the adapters in both of my wired PCs. That surprised me because my benchmark tests were all consistent with 100baseT, not 1000baseT. I researched that and found that one system was running at 1000 Mb/s and the other at 100 Mb/s as reported by ethtool. I fiddled a bit trying to get ethtool and mii-tool to change link speed to 1000 Gb/s without success. I tried a different cable and that made the difference. However, the best TCP throughput I found using netperf is about 300-500 Kb/s. I'm not sure what causes that discrepancy. Perhaps cabling is still an issue since I may not not be using Cat 5e. I'll revisit this when I start getting upgraded systems in place. H/W for the symbiotic NAS should be here soon. (And I'm jonesing for an upgrade to the desktop. I've already picked out the H/W. :D I've just discovered that the present system is not actually capable to play a DVD smoothly!)

No comments:

Post a Comment