View Full Version : Near/Far is killing me.
Skaught
03-22-2006, 01:26 PM
I have a few sectors with huge near/far issues.
In one example the weakest client has -67 and has 80% packet loss on 1300 byte pings. Clients better than -65 are fine. I have ruled out channel issues.
Is there any way to counter near/far in 802.11? I already have the near clients backed off as far as the chipset will let me.
HoeDing
03-22-2006, 04:05 PM
Have you tried attenuators on your near antenna's? (I personally have never tried this, I am just throwing it out there)
Skaught
03-23-2006, 12:01 PM
The near antennas are intergated CPEs and already are -12db padded. Adding an anntenuator would mean destroying them to get inside.
bobbyc
03-23-2006, 05:11 PM
Skaught,
Are the APs prism unamped? -67 should be fine. Do you feel that the strong signals of the near clients are somehow destroying the sensitivity of the AP?
Or do you think that it is a timing/beacon mismatch issue?
If the former, you might have to visit each cpe with a strong signal and give them downtilt.
If the latter, I'm not sure what the solution would be.
An interesting idea you could try if you still have problems after option A is to try logging into each CPE and setting the RTS and Fragmentation numbers back to default. We've always left RTS and Fragmentation as large as they go (I think 2346 or 2347) and haven't had much problems.
I might be experiencing the same problem as you on one of our towers, but I've yet to start the troubleshooting since the packetloss isn't nearly as bad as what you're seeing... the AP is a amped orinoco card, and it is getting switched to a non amped tower mounted atheros AP very soon. If the problem exists after the swapout, I will probably start by visiting the nearby customers and downtilting their CPE.
Bob C
Skaught
03-24-2006, 04:10 AM
IT is a 100% prisim system.
all the clients have 97% or better uptime (most 100%) but if I look at how much they have good service, those who are far away are as bad as 6% good. Not many are that bad and boosting their signal helps.
It is really bad when the guy -52 starts downloading something.
I can say I have plenty of examples of setups like this where I have 10 - 40 people on an AP, a few have -40 and -50 and many have -70 to -80. I've never had any problems. I do run all atheros.
Skaught
03-24-2006, 12:28 PM
This AP has 97 clients on it. On many of my towers I have run out of channels and there is simply no more room left. This one tower has 258 clients on it.
I am building a 5ghz ptmp product but it is a few months off.
97 clients on one AP! Holy balls!
You can't possibly be selling 1.5+ mbit connectivity to the users and managing that density...
Skaught
03-27-2006, 05:46 PM
Actually for 95% of the clients on the AP they get >2mbit 20 hours a day and more than 1 the other 4 hours a day.
I was never able to go this high until I turned on RTS.
That site has 3 sectors and they are all full. Hence our need to go to 5ghz. We have simply run out of channels.
We also have very strict standards for signal checks and 100% of our CPEs are CB3/CPE200 based.
I think having a mismash of CPEs really kills performance. Certainly the CPQs did.
Skaught
03-29-2006, 03:40 PM
So far reducing the RTS on my cpes yet lower has helped dramatically. I have only tried this for about a day and a half though but in some cases I saw an INSTANT dramatic improvement.
ninedd
03-30-2006, 01:04 PM
Skaught, what settings are you using for RTS/CTS?
bobbyc
03-30-2006, 03:10 PM
About 20% of our CPE do not support RTS/Fragmentation... so we leave it disabled on all CPE. The LIMITED testing I've done showed no improvement with RTS or Frag. set on the CPEs we have that support it. It actually slowed down the max rx/tx speeds on the AP.
Again, we've done LIMITED testing, and never with a busy AP.
ninedd
03-30-2006, 04:08 PM
For us, other than a one SmartBridge and one WET-11 that are still out there from years and years ago, we've standardized all the CPE's to support RTS, FRAG & Routing in the CPE to keep the customer's computer(s) off our network.
The logic of RTS seems sound - with the lack of a polling mechanism in 802.11x, it's the next best thing. The Radio checks if it's clear to send before actually sending. That should help stop people from bumping into each other as much, especially on uploading. We currently set all CPE's to 768 RTS size, and it does seem to slow things down a bit. However, blazing top speed has never been my #1 priority - it's getting everyone to work more consistently.
For FRAG, we've never seen it help / hurt anything. We tried it several times on marginal systems. Someone that we can ping 100 out of 100 with a 512byte packet but that has packet loss at 1472, we've figured that maybe FRAG was supposed to break those large 1472's into multiple 512's. However, we've never seen it able to do anything, and not sure why. Not a big deal, since the real soluion is to fix the link so that it can do 1472's anway. :)
So, I'd love to hear what other people's experience with RTS / FRAG is?
go.fast
03-30-2006, 04:22 PM
Does the war v3 suport rts?
I can't seem to find it.
George
Skaught
03-31-2006, 02:25 PM
Skaught, what settings are you using for RTS/CTS?
t the moment. 1 I was on 150. The majority of my packets are well under 50 bytes and without it the collisions are HUGE.
It seems most packets in limewire are under 50 bytes, dropping the rts has virtually eliminated my limewire issues and limewire still works.
Speeds have gone up across the board, I have done tests giving me 400KB/s on my AP with 97 clients on it on a http download.
Skaught
04-01-2006, 01:03 AM
I now have enough quantitative data to release an early but official result.
I had my office intern spend 2 days this week and go into every single CPE on my network and move the RTS from 150 to 1. And the change is amazing.
In my daily charts of latency/throughput, My usual spikes of bad latency in the evenings has essentially vanished. Also my near/far problems have almost completly vanished. I have a few clients that start dropping about 30% of 1300 byte pings at peak time but they are very rare now.
Plus I have about 2% of the CPEs where the web interface on them locked up so they are still on the old RTS numbers. Once they get switched I may see an additional jump in performance.
And as far as overhead I have not seen it. One of my staff lives on my AP with 98 subscribers and he got 400KBytes/s at 7pm yesterday no problem and that was the first AP to get switched to an RTS of 1.
I have logs over the last week that show quantitatively it makes a big difference on heaviliy loaded sectors, near/far and hidden node.
Wow, thanks for the results.
The little bit of messing around I did with enabling RTS/CTS didn't go well for me, I always saw huge reductions in upload and download performance and nasty latency.
I think part of my problem may be that I didn't go switch every single client on the AP to RTS/CTS though, only a couple, and those couple suffered because nobody else on the AP was being as polite as them.
Thanks for sharing your results. Now I am rethinking about rts/cts option as we have much of upload traffic in our network.