Live Music Archive
Librivox Free Audio
Metropolitan Museum
Cleveland Museum of Art
Internet Arcade
Console Living Room
Books to Borrow
Open Library
TV News
Understanding 9/11View Post [edit]
| Poster: | steffenyount | Date: | Oct 31, 2003 7:27am |
| Forum: | sflan | Subject: | Wired-link throughput much higher |
http://woody.archive.org/sflprobe/throughput.cgi
and testing various link throughputs.
I was a rather suprised to see the results showing many of the wireless link throughputs as being around 1Mb/s, much lower than 11Mb/s. E.g. sflan14 to sflan11 ~800-900Kb/s
While in comparison the wired links were performing with better than 11Mb/s throughput. E.g. sflan7 to sflan14 ~14-19Mb/s
Why is the wireless throughput so low?
Is it because of the testing program being used?Because of the physical distance? Because of the antenna's being used? Because of 2.4Ghz band traffic interference? Or because of the wireless NICs being used?
Has this been investigated?
Kind Regards,
-Steffen
Reply [edit]
| Poster: | TZapper | Date: | Nov 1, 2003 2:16pm |
| Forum: | sflan | Subject: | Re: Wired-link throughput much higher |
Reply [edit]
| Poster: | steffenyount | Date: | Nov 1, 2003 3:38pm |
| Forum: | sflan | Subject: | Re: Wired-link throughput much higher |
Should I expect the same kind of performance from my possible ~2 mile links towards downtown?
How does 802.11g compare in terms of throughput under the same conditions?
What can be done to reduce the interference?
High gain directional antennas?
Antenna filters?
Cheers,
-S
Reply [edit]
| Poster: | my13 | Date: | Nov 2, 2003 2:29am |
| Forum: | sflan | Subject: | Re: Wired-link throughput much higher |
First, we're only just beginning to look at performance issues. The figures you're seeing are reported by the iperf tool, which measures bytes you can send over a TCP/IP connection. It does not count the TCP/IP overhead, including the burden of the returning ACKs (and 802.11b is half duplex). Therefore, the throughput reported by iperf is somewhat less than the link-level capacity. Thus, if you run the same test over, say, a 384 Kb/s DSL line, you will get less than 384 Kb/s reported throughput.
A more serious limitation in the current test is that it does not account for contention with other traffic on the link: it just measures how much data the test itself can transfer. If others are using the link at the time of the test, reported throughput will be lower. We'll definitely be putting some effort into enhancing the measurement tools.
However, under the best of circumstances, you'll never get close to 11 Mb/s of actual throughput. See http://lists.linux-wlan.com/pipermail/linux-wlan-user/2003-February/009337.html, for example, for a discussion of some of the overhead. In practice, I doubt we'll ever see more than 5 Mb/s from the iperf test; we're currently close to that on the link between sflan1 and sflan2.
With the extra distance between sflan nodes, I'd be very happy if we consistently say 3 Mb/s between all links. We're definitely not there yet today.
Why not? You mention a pretty good "laundry list" of possible reasons. One that you didn't mention is interference between the two radios on the nodes. We're certainly interested in learning more about how these different factors conspire to reduce our effective throughput.
In the long run, things look brighter. New modulation techniques, maybe even new frequency bands, and higher node density should increase throughput up into the 10-100 Mb/s range. Whether that will be on proprietary networks operated by cell phone companies, or on open, community networks remains to be seen.
Credit where credit is due: much of what I've written above is a distillation of conversations with John Berry, who sometimes posts to this forum. Any gross inaccuracies are certainly a distillation error on my part :-).
