Slow Lane On The Information Superhighway?

[Brad Adrian]

<div class='os_post_top_link'><a href='http://www.pcworld.com/news/article/0,aid,111834,tk,dn073103X,00.asp' target='_blank'>http://www.pcworld.com/news/article...n073103X,00.asp</a><br /><br /></div>In a paper presented at the Infocom 2003 conference, French researchers say they have identified a problem with standard WLANs that can severely slow down the overall speedof the network.<br /><br />"Engineering experts at France's Centre Nationale de la Recherche Scientifique (CNRS) in Paris have demonstrated how a slow device connected to WLAN hotspot can cause an entire network to drop to a slower speed, citing a common channel access method as the cause...In other words, when a device with a low bit rate captures the channel, it penalizes other devices using a higher rate by degrading the speed of their connections."<br /><br />Somebody needs to explain the technical details of their findings to me, but it sounds to me like they're saying that a slow or older wireless device on the network can have the same effect as getting stuck on the autobahn behind a 1956 DeSoto doing 25 mph. The important implication, of course, is that WLANs used by disparate devices, like public WiFi hotspots (or even corporate networks) can be subjected to troublesome slowdowns.<br /><br />I sure hope that this isn't as bad as it sounds. If word spreads that connectivity can be hampered by a single slug on the system, this could make it even more difficult for WiFi hotspots to gain adoption and earn any real money.

[phanprod]

I believe (and I could be wrong) that the idea works like this: Connect an 802.11b device to a 'g' access point, and you're down to 'b' speeds across the board. It's the same with most ethernet hubs, where connecting a 10Mbs device to a 10/100 hub will slow the entire network down to the 10Mbs speed.

It makes sense, considering that you'd need to have two discrete wireless tranceivers to theoretically operate two protocalls at one time. I'm sure that someone will come out with a wireless device that acts more like an ethernet 'switch' or 'router', alternating protocalls on a per-data packet basis to deal with the issue... if it doesn't exist already.

However, for most internet purposes, 11Mbs is a pretty decent speed - considering it's faster than most xDSL lines... and, with multiple users connected, the available bandwidth decreases regardless. So you wouldn't get full speed on a crowded public access point anyway.

[Ed Hansberry]

Oh great. That's all we need. Someone with a 66MHz Palm device and a SD WiFi card clogging up the system. :rotfl:

[mr_Ray]

I believe (and I could be wrong) that the idea works like this: Connect an 802.11b device to a 'g' access point, and you're down to 'b' speeds across the board. It's the same with most ethernet hubs, where connecting a 10Mbs device to a 10/100 hub will slow the entire network down to the 10Mbs speed.

It makes sense, considering that you'd need to have two discrete wireless tranceivers to theoretically operate two protocalls at one time. I'm sure that someone will come out with a wireless device that acts more like an ethernet 'switch' or 'router', alternating protocalls on a per-data packet basis to deal with the issue... if it doesn't exist already.

However, for most internet purposes, 11Mbs is a pretty decent speed - considering it's faster than most xDSL lines... and, with multiple users connected, the available bandwidth decreases regardless. So you wouldn't get full speed on a crowded public access point anyway.

If this is the 802.11g/b issue (the link isn't working for me so I can't check) then this has been known for some time. - All devices connecting to the same point need to use the same version (a/b/g/whatever). At least one manufacturer is working on a way round this though and we'll probably see more flexible base stations next year. Also I think some devices allow you to force backwards compatibility off, too.

I've also heard of problems with bad/slow devices slowing the whole system down, but I must admit to not knowing the details there or if it's possible to fix.

[MarkWorsfold]

If you are driving a Skoda on the M25 you will be luck to reach 25 mph!

[Pony99CA]

Oh great. That's all we need. Someone with a 66MHz Palm device and a SD WiFi card clogging up the system. :rotfl:
:lol: Unfortunately (or fortunately), I doubt processor speed has anything to do with it; it's the network speed only. If that 66 MHz Palm had an 802.11g SD card (if and when those exist), the 400 MHz Pocket PC with the 802.11b card would slow it down.

I wondered why a slow-down would occur, and it didn't make sense. Then I thought about it.

If my understanding of networking is correct, when a device sends a packet out, all devices on the network see it, but only the one that it's addressed to will actually process it. However, all devices need to be able to check the TCP/IP header to see if it's for them.

If all of the devices on the network work at the same speed, there aren't any problems. However, if a slower device connects, if the network kept sending data at its maximum speed, the slower device probably wouldn't be able to read its packets. So, in order to allow that device to see its packets, all traffic has to slow down.

Not being a networking guy, though, that's just a guess.

Steve

[Ed Hansberry]

Oh great. That's all we need. Someone with a 66MHz Palm device and a SD WiFi card clogging up the system. :rotfl:
:lol: Unfortunately (or fortunately), I doubt processor speed has anything to do with it; it's the network speed only.
I remember reading that the Handspring Visor with an 802.11b springboard card was maxed out at around 700bps because the processor simply couldn't handle data coming in any faster.

[racerx]

If the processor can handle the "bandwidth" of the wireless NIC, then the processor wouldn't slow it down. But yes, you could get an older Palm by way of an Ethernet sled that could slow things down.

And an 11MB (actually around 6MB of actual throughput - rest is communications overhead) will sow down a 802.11g access point. You need devices all running at the same level in order to get the maxium throughput (that includes the access point). This is one reason why hubs have all but disappeard from the wired Ethernet world. Everyone uses switches now for reasons like this. Unfortunately, your access point acts like a hub, not a switch, even if it has a switch built-in to it.

[Dave Beauvais]

... If my understanding of networking is correct, when a device sends a packet out, all devices on the network see it, but only the one that it's addressed to will actually process it. However, all devices need to be able to check the TCP/IP header to see if it's for them. ...
In a non-switched network, that is correct. This is why hubs suck and switches rock. :) Switches have an internal buffer that does what's called "store and forward" that buffers the data intended for each port so that the attached device can be fed the data at whatever rate it can accept it. That's why you can have a 10/100 Mbps switch with both 10 Mbps and 100 Mbps devices attached with no decrease in performance. If the slower devices are especially slow, however, even with the data buffer in the switch, some data will likely need to be retransmitted since the buffer will eventually overflow.

As far as a know, all current wireless access points are essentially wireless hubs, not switches, which is why throughput drops when a slower device connects.

--Dave

[othell]

Part 1:
http://www.tomshardware.com/network/20030310/index.html

Part 2:
http://www17.tomshardware.com/network/20030317/wireless_ntk2-01.html

Could both of these articles answer your questions? You gotta read all of them though to fully understand much of what's said. Plus it'd be good just to know it....

[hawkeye]

That's why I won't buy a PDA with wi-fi built-in. Just having my PDA on my G-only home network would kill my bandwidth. I'll buy a PDA w/wi-fi or a Linksys Media adapter when they offer them with G-wireless.

[Brad Adrian]

Wow! I never knew there were so many people in this community who know so much about so many things!

Thanks for explaining this to me in terms I can understand. It worries me, though, that this kind of problem might lead to a whole new form of discrimination. Will hotspots be displaying signs like "No Bs allowed!" or will they have nice lounges for the G users and crummy ones for the Bs?

[Dave Beauvais]

... It worries me, though, that this kind of problem might lead to a whole new form of discrimination. ...
I doubt that'll happen anytime soon. Why? Because I doubt most people give a crap. :) All they want to do is a working connection to be able to surf and check e-mail while sitting in a cafe. Even "slow" 802.11b is more than sufficient for these uses.

--Dave

[Howard2k]

... If my understanding of networking is correct, when a device sends a packet out, all devices on the network see it, but only the one that it's addressed to will actually process it. However, all devices need to be able to check the TCP/IP header to see if it's for them. ...
In a non-switched network, that is correct. This is why hubs suck and switches rock. :) Switches have an internal buffer that does what's called "store and forward" that buffers the data intended for each port so that the attached device can be fed the data at whatever rate it can accept it. That's why you can have a 10/100 Mbps switch with both 10 Mbps and 100 Mbps devices attached with no decrease in performance. If the slower devices are especially slow, however, even with the data buffer in the switch, some data will likely need to be retransmitted since the buffer will eventually overflow.




As far as a know, all current wireless access points are essentially wireless hubs, not switches, which is why throughput drops when a slower device connects.

--Dave

Very close. The WLAN is a shared media like a hub, but the devices do the lookup at layer 2 (the MAC address). IP is layer 3 and TCP is layer 4.

Secondly while one device is transmitting another can not. So a slow device that takes too long to send data will slow down other devices while they wait for it to finish.

If you have multiple PCs, have all but one set for 11Mb/s, the other set a 1Mb/s then the other will be disproportionatly affected by the 1Mb/s device.

[Jer]

Will an A/B router suffer the same ill effects?

[Dave Beauvais]

Will an A/B router suffer the same ill effects?
I can't say with 100% certainty that it won't, but I will say that it shouldn't. Because they operate on different frequencies, 802.11b and 802.11a must use separate radio transceivers. As a result, the slower 802.11b devices shouldn't have much of an impact on the faster 802.11a devices. However, I guess it would depend on how the manufacturer designs the hardware and firmware that allows the two radios to communicate internally.

Honestly, I think 802.11a is going to die out pretty quickly now that 802.11g has been ratified. About the only thing 802.11a has going for it is its relative immunity to interference from things like microwave ovens and cordless telephones that share the 2.4 GHz frequency range used by 802.11b and g.

--Dave

[Howard2k]

But a g router in mixed mode will suffer similar effects... Well documented though..

[Pony99CA]

Honestly, I think 802.11a is going to die out pretty quickly now that 802.11g has been ratified. About the only thing 802.11a has going for it is its relative immunity to interference from things like microwave ovens and cordless telephones that share the 2.4 GHz frequency range used by 802.11b and g.
There are companies producing a/b/g (or a/g) cards that don't seem to agree with you. Another (unintended) benefit of 802.11a is that it is more secure than 802.11b due to fewer people having 802.11a equipment and because it is not compatible with 802.11 b/g. Very few bozos with a war driving rig will be able to find 802.11a networks.

Yes, that's like security by obscurity, but unless a/b/g cards take off, it will keep a lot of people out.

Steve

[ctmagnus]

Very few bozos with a war driving rig will be able to find 802.11a networks.

Just wondering: Do they even make non-Cardbus (ie, 16-bit) 802.11a cards? Cuz if not, then you can't get 802.11a on a Pocket PC.

Also, apparently some 802.11a cards will function with NetStumbler.