WRT54GS WDS bridge bandwidth problem

Hi, I setup a WDS bridge between two WRT54GS wireless routers running Alchemy 6.0rc4a, I have a laptop connected (wired) to the LAN 1 port of the first WRT and another one connected to the LAN 1 port of the other WRT.

I put the WRTs pretty close: according to the wireless status page, they are running a full 54 Mbps link with a RSSI of about -35 dBm.

I ran some bandwidth tests for the wireless bridge using the 'iperf' utility:

I only got about 12 Mbps, I tried both ways: from laptop 1 to laptop 2 and viceversa, the results are about the same.

I think I should get a lot more bandwidth (about the double, e.g. 24 Mbps) out of a full 54 Mbps link.

Any idea?

What are your results?

Thanks a lot.

The problem here is that you plugged everything together and found that it didn't quite work. The trick is to find which *PART* of the problem isn't working as expected. Some things to try:

1. Please disclose the exact command line and non-paraphrased output from IPerf. What buffer sizes did you use? Were you testing TCP or UDP? (UDP is somewhat faster). Between what two operating systems (Linux, Unix, Windoze, etc)?
2. It would be interesting to know what IPerf reports without the wireless link. Can you move one of the computers next to the other and see what IPerf reports with a direct LAN connection?
3. Do your WRT54GS statistics show any wireless errors or resends? It doesn't take much RF interference or reflections to drop the error rate. Were the results consistant? Try it with *LESS* RF signal by disconnecting both antennas on one end.
4. Are your ethernet ports on the computahs running at full duplex? With 100baseTX-HDX, the best I could do with a direct connection (on a Windoze 2000) machine was about 60Mbit/sec using large packets.
5. Anything connected between the WRT54GS boxes that might slow things down such as a 10/100 Hub?
6. Are you running any form of encryption (WEP or WPA)? Depending on implimentation, this will slow you down somewhat.
7. I'll confess that I don't really understand how WDS works. By that, I mean how many packets are shoveled back and forth, or whether there is or isn't a performance penalty for using WDS as a point to point bridge. Alchemy supports a client mode. Put one radio in the access point mode, the other in the client mode. Try again and see if there's a difference.

Actually I don't have any wireless repeater in the middle, I just have two WRT54GS wireless routers setup in LAN WDS bridging mode.

No other wireless equipment, the laptops are connected (wired) to the LAN ports of the WRT54GS boxes.

Any idea?

Thanks.

FWIW, a not-very-scientific data point follows.

Using a Buffalo WBR2-G54 (AP/router) and a WLA-G54 (AP) linked by WDS (client access enabled on the router, but no active wireless clients at the time of the test) with a good strong 802.11g signal and S/N ratio, I moved more than 3GB from the WLA-G54 to the WBR2-G54 at a fairly steady rate of 24 to 25Mbps. (I couldn't really measure the wireless transfer rate, so that figure represents what was going out from the sending computer via Ethernet.)

I can't give you comparable figures for a client connected to either of the Buffalo boxes, because I don't actually have any 802.11g client devices.

Ok. Those are fast enough to do a decent test.

Yes, it's TCP by default. Default's the problem. IPerf starts with a very small buffer size (50KB?) which will be REALLY slow. Try:

Server Side: iperf -s -w 500k Client Side: iperf -c -w 500k

500k is probably too large, but it's what I use for a first approximation.

Also, be sure to run: iperf -s -m and see if it complains about MTU discovery failing. What's the MSS (or MTU) size reported? If it's stuck with small packets, it's gonna be slow.

Do you really expect a specific answer without supplying any details? Try answering some of the questions in this newsgroup and you'll notice that over half the posted questions don't bother to supply any numbers or details. Apparently, it's a common expectation.

Ok, then it probably wasn't RF interference as that tends to change in level, intensity, effects, and such between tests.

Good. 90Mbits/sec is about wire speed for 100baseTX which requires full duplex to get that speed. Both machines and Iperf are apparently working correctly.

I haven't the foggiest idea. There are support groups for Sveasoft firmware that will have better clues on using their tools. Usually, there's some type of status page with includes MAC level errors. If Alchemy supports SNMP, it can usually be extracted from there.

I've noticed on the bench that there is a miniumum seperation between radios, but have never investigated the cause. It could be overload, timing, or my imagination. Usually about 1 or 2 meters seperation is sufficient to eliminate any such effects. If the units were next to each other, there *MAY* be problems.

When you get it going faster, it would be interesting to know how much WPA slows down the thruput.

Well, the idea was to eliminate WDS from the performance picture. The reason I listed it last is that it's the test that will burn the most time. However, judgeing by the numbers offered by Neill Massello seem to indicate that WDS does not present any overhead problems. You should get full speed. Save the non-WDS tests for last.

One other dumb test worth trying. Just copy a large file with ftp and time the transfer with a watch. Do the math and see what speed you get without using Iperf.

The first laptop (Dell Inspiron 8000 - P3 1Ghz - 512MB RAM) is running Windows XP Professional SP2, the other laptop (Apple PowerBook 17" - G4

1.5Ghz - 2GB RAM) is running Mac OS X Panther (10.3.7). There is no firewall software enabled on the laptops.

I ran the utility in server mode on one laptop (iperf -s) and in client mode on the other one (iperf -c 192.168.0.x). I didn't use any special option (except for running the test for more seconds than the default setting). I think the utility does a TCP test by default, using the default OS TCP window size.

I don't have the iperf output handy now, but it was about 12Mbps. I did serveral tests, in both ways: from laptop 1 to laptop 2 and from laptop

2 to laptop 1. The results were about the same.

I tried that too: I connected both laptops to the LAN ports of one WRT54GS box, and IIRC I got about 90Mbps, or a little less.

How do I check the wireless errors and resends in my WRT54GS? I'll try to remove both antennas on one end and let you know if it helps.

Yes, the laptops and WRT54GS LAN ports are running 100Mbps full-duplex.

Negative, the laptops are connected directly to the WRT54GS boxes, and the WRT54GS boxes are connected directly with a wireless WDS bridge.

Negative, there is no encryption on the wireless link.

I'll try that and let you know, but I think that the WDS bridging feature is the recommended one for a wireless bridge by the Sveasoft people. I am not sure if Alchemy client mode suffers of the same limitation of Satori: only one computer can be wired to the WRT54GS LAN.

Thanks.

24 to 25Mbps, that's a good throughput, that's what I would expect from my WRT54GS boxes too ...

Did you use any special option for your WDS bridge?

Thanks.

Frame bursting is enabled, and the bridge unit is doing WDS exclusively ("dedicated mode" in Buffalospeak). Nothing else special, but I should also have mentioned that the WDS link uses 128-bit WEP.

I already enabled the frame bursting option on my WRT54GS boxes, it didn't make any noticeable difference ...

I couldn't find any WDS 'dedicated mode' in the Sveasoft firmware.

What radio chipset do you APs use?

Thanks.

I'll try it asap and let you know.

You are right. The problem is that I don't have the LinkSys WRT54GS boxes next to me, there are in another location, I cannot do other tests until this week-end.

Okay, I am going to do some 'googling' for that.

There separation between the WRT54GS boxes is about 2 meters and the RSSI is about -35 dBm (overload shouldn't be an issue).

I am going to try more separation and/or to remove the antennas from one AP, just to be sure.

Yes, I plan to turn on WPA as soon as I get a good bandwidth without encryption. I'll let you know how it affects the throughput.

Agreed, I'll save the AP/Client test for last.

I already tried a similar test: I got about 1.5MBytes/sec trasferring a file via SMB/CIFS (Windows sharing). That is about the same throughput reported by the iperf utility: 12Mbits/sec.

Thanks.

I'm pretty sure Buffalo uses Broadcom in most of their wireless products.

That's interesting, since the WRT54GS wireless router also a Broadcome chipset. In theory I should get the same (or similar) throughput of your WDS bridge ...

I tried a bigger window size (about 200K, since it didn't accept 500k), it didn't make any noticeable difference from the default setting.

Here is the iperf output from the MTU discovery test: