Wi-Fi Overload at High-Tech Meetings | Hacker News Reader

Wi-Fi Overload at High-Tech Meetings

nytimes.com

26 pointswallflower19 comments

At Defcon every year, we have ~10k attendees, with literally thousands of DHCP leases active, using a combination of Aruba wireless gear, and Cisco switches. We don't generally have a problem.

Good planning is key; we have heatmaps (signal strength predictions) for every venue in the conference, careful AP placement, and a butt-load of security backend to keep things running smoothly, combined with some EXTREMELY talented staff running the configurations and monitoring continuously over the weekend. The network staff shows up a WEEK in advance, and often does a preflight in April to check on any changes in the hotel (we are switching hotels this year, so there is a lot more prep this time around). Not many conferences have a dozen high pay grade network admins prepping for a full week in advance. But you cannot argue with results.

The real choke point is probably the 100Mbit link back to the ISP, but we keep upping it every year ;)

See here: http://www.wired.com/epicenter/2010/12/ruckus-wifi-3g/

Ruckus set up the access points MacTech Conference in Nov, I was seriously impressed.

At most conferences I've been at, the wifi tends to die or have performance problems.

In this case, it was actually usable with effectively no performance/signal issues even with 200+ people (probably 300+ devices between laptops/iPads/iPhones) in a large hotel conference room.

Joel Spolksy posted about this problem on his blog last year: http://www.joelonsoftware.com/items/2009/10/08.html

Here's the related ServerFault post: http://www.joelonsoftware.com/items/2009/10/08.html

Correction: ServerFault discussion here... http://serverfault.com/questions/72767/why-is-internet-acces...

So many people, even people who work @ Apple in the Airport group, don't understand wifi.

Here is one anti-pattern for WiFi... run the APs co-channel. run them all on, say, channel 6.

Because in the 2.4GHz ISM band, given OFDM modulation, and the minimum standards set by the IEEE, there is really only one usable channel.

The adjacent channel interference problem is worse with 802.11n receivers, btw.

But if you run the APs co-channel, then CCA can work.

> there is really only one usable channel.

Can you elaborate on this a bit? I'm running all 802.11n hardware for my personal networks, but I troubleshoot g networks from time to time. I've never heard this assertion before.

I had to convince a former bosses boss of mine that this was the proper setup. He had setup the network when the company was tiny and explaining what needed to be changed when we grew to more than 10 times the users and 20 times the wifi devices was a lesson in convincing a brick wall. Getting all the best practices collected was not an easy task.

Ugh, that's a piece of non-news if I ever heard one. NYT: Wi-Fi Slow at Conferences.

Is there a point to this article, or are they just stating something that no one cares about?

It is most definitely not news, but going so far as to say it's something no one cares about is going a little far.

Perhaps someone more technologically inclined than I am can shed light on this but when a company like Apple hosts a marquee event like WWDC why can't they see how many people will be at the event, then deploy enough wifi routers to meet demand? Especially when Apple makes their own wifi routers. Across wifi a,b,g,and n there should be plenty of spectrum to go around. If a company can deploy 10,000 servers in one room, why can't they deploy 100 wifi routers?

Do you have the manpower and budget to get those 100 wi-fi routers set up and tested in 1 day? Most convention locations don't allow for a huge amount of prep time.

It can be done if the people have good equipement and REALLY know what they are doing.

The trouble is that the venue hire the cheapest people to fit the system and have no clue how to run it. Since nobody knows how the good the system is before they do the show you have to trust the venue - when you find out it's crap it's too late - so there's no incentive to do a good job.

Because there aren't 100 separate wifi channels (frequencies) so if you put big powerful routers with lots of signal then people from the other end of the room will try and connect, if you put little low power ones then they intefere with each other.

Then you have the overhead of 1000s of people trying to connect, their wifi cards not receiving a response because the router is busy so they try on another channel tying that one up as well.

Wifi really isn't designed for this sort of setup - it's amazingly difficult to get it right, throwing hardware just makes it worse

You can install as many of the highest quality routers as you like, but you can't control the devices. The devices interfere with each other because they weren't designed for or tested in an environment with thousands in the same room (and many if not most WiFi devices have crappy hardware and/or software).

Intel wrote a whitepaper[1] on this very subject. One of the more amusingly counterintuitive findings is that you want each radio to cover not a whole lot of area. No matter how fast the backhaul is, the spectrum can only handle a couple dozen clients, which fills up fast when you've got tens of thousands of attendees.

1: http://www.intel.com/it/pdf/conference-wifi.pdf

Could work: a combination of wired connections, asking attendees to bring their own 3G dongles and providing a local cell base station for areas with poor mobile coverage.

For people with ethernet ports, you could spread a few wifi routers with wired ports, so wifi traffic is carried over to the fewer routers than to each individual client.

This isn't just a wifi problem it is a general wireless problem. Where there are large gatherings like The Daily Shows Rally to Restore Sanity for example cell service totally failed in the area around the washington mall.

You can't solve the problem by just increasing the # of routers, It's the noise and the lack of frequency in the space that causes the issue with so many nodes.

This is a serious problem for a network I am designing. With many many systems on the wireless network we are ensuring that we have better coverage with WAPs - but the real problem is because we will have more than 10,000 wifi Aeroscout tags on the network.

Both Cisco and Aeroscout have no idea what this level of activity will look like. While Aeroscout's largest client has some 30,000 tags - they are on a much larger square footage campus.

I am still trying to find the best approach to this problem, but any of the companies that make the HW are only interested in selling you more hardware - not developing a solution, as of yet.

The network is more than $15M in HW alone - so, I am sure I will bring them around.

Reply on news.ycombinator.com