Follow Slashdot blog updates by subscribing to our blog RSS feed


Forgot your password?
Communications Python Wireless Networking

Why PyCon 2010's Conference Wi-Fi Didn't Melt Down 145

jafo writes "There's been a lot of teeth gnashing going on recently about broken wireless at conferences. We just wrapped up PyCon 2010, with around 600 (out of 1,000) attendees simultaneously accessing the volunteer-run network, and response has been fairly positive. 2.4GHz (802.11b/g) continues to be problematic, but most users were on 5.2GHz (using 802.11n) and associating at 130mbps, with a 100mbps link to the net (though after the fact we found that 35mbps would have sufficed). My PyCon 2010 wrap-up reveals all the secrets of how we did it, including pretty bandwidth and user graphs."
This discussion has been archived. No new comments can be posted.

Why PyCon 2010's Conference Wi-Fi Didn't Melt Down

Comments Filter:
  • Editors, please do your job before you accept a story - that's an easy way to make Slashdot much better. In this particular story, it would have been easy - no research required. As I'm sure almost everyone here knows, m != M. Also, what is wrong with "b/s" instead of "bps"? (Also, how do I write non-ASCII characters here?)

    • by Jurily ( 900488 ) <jurily@g m a i l . com> on Thursday March 04, 2010 @09:58AM (#31357422)

      (Also, how do I write non-ASCII characters here?)

      What makes you think you can, huh? Just because url's can contain them and the civilized portion of the net is already fine with them?

    • Pet peeve of mine. Big one. It's really hard for me to take someone seriously who
      writes about millibits while meaning megabits.
    • Re: (Score:2, Informative)

      For some, like ä, ß, €, you can just use the HTML entities (&auml;, &szlig;, &euro;). More esoteric ones like &#x0950; just won't work.

    • by Rogerborg ( 306625 ) on Thursday March 04, 2010 @10:29AM (#31357698) Homepage

      Just FYI, the "job" of a Slashdot "editor" involves scoring Rob Malda some weed when you were at community college together, writing a very small shell script to post every 25th story submission, then scarfing beer and cheetos while playing in the Furry zone of Second Life for the rest of your "career".

      Mod hints: -1 Troll, +1 Informative, +1 Insightful

    • Most books on computer networks that I have seen use bps. But I haven't seen many. I suppose the counter question would then be: What's wrong with "bps" instead of "b/s"?

      • by TeknoHog ( 164938 ) on Thursday March 04, 2010 @11:28AM (#31358234) Homepage Journal

        What's wrong with "bps" instead of "b/s"?

        "Per" is a word that does not work in all languages, whereas "/" is a universal mathematical symbol. Even non-scientists use units like km/h, at least in Europe.

        • Re: (Score:3, Insightful)

          by trapnest ( 1608791 )
          I suppose that would be really important if the summery were not in english.
        • by jadin ( 65295 )

          Is "per" even an English word originally? Looks like something we pillaged from Latin or wherever.

      • by rdnetto ( 955205 )

        p is SI prefix for pico. So bps could be interpreted as bit-picoseconds.

        • by ls671 ( 1122017 )

          yeah but: bits * picosecond (bps)

          doesn't make much sense, it would need to be bits / picosecond.

          • by rdnetto ( 955205 )

            It's common to have Nm (Newton metres aka. Joules) and kgm/s when working with SI units. So bs (bit seconds) could be a unit, regardless of whether or not it has any use.
            As for bps = bits/picosecond, you'd need 2 ps for that, which only emphasizes its ambiguity. Besides, basic algebra states that 'xy=x*y' and 'x/y = x÷y', so it wouldn't make any sense.

            • by ls671 ( 1122017 )

              > It's common to have Nm (Newton metres aka. Joules) and
              > kgm/s when working with SI units. So bs (bit seconds)
              > could be a unit, regardless of whether or not it has any use.

              I understand all this, I meant not much sense in the context we were speaking about, for bandwidth usage and even then. And context is important, even in science.

              Units like Kilowatt * hour, kW-h make sense because you end up with a realistic unit once we simplify the units used: 1 kWh = 3.6 megajoules.

              Units for bandwidth have

              • by rdnetto ( 955205 )

                Sorry but I still can't envision what a bit * picosecond could represent in the context of bandwidth.

                I never said it had to represent something, only that under the accepted notation it would mean something completely different.

    • Thank you. I agree completely. Units are important and are not just letters to be applied willy-nilly. The m and M are not lower and upper case versions of a letter they are completely different symbols.

    • (Also, how do I write non-ASCII characters here?)

      Only the printable ascii table and extended ascii table seem to be supported. If the HTML ascii code (e.g. é=&#233;) is greater than 255 (a.k.a. ÿ), then you can't get there from here. This fact has been pointed out by many users... no unicode, no change.

      • by ls671 ( 1122017 )

        > If the HTML ascii code (e.g. é=&#233;) is greater than 255...

        It should have been: "If the HTML code (e.g. é=&#233;) is greater than 255"...

        ASCII codes can't be greater than 255 and this is extended ASCII ;-)) []

        I realize that you already seem to understand this, you probably just made a mistake typing... ;-)

    • There are some HTML entities that work. &mdash; = [—], &rdquo; = [”], &ldquo; = [“], &lsquo; = [‘], &rsquo; = [’], and so on. But I don’t know which ones are allowed, and which ones not. &hellip; for example, is not allowed: []. Even hexadecimal ones only work when allowed as normal entities.
      I haven’t found much use for others here. Although I wish, mathematical symbols would work. As I have a ton of them on my keyboard layout [].

  • Typo in model number (Score:3, Informative)

    by madsci1016 ( 1111233 ) * on Thursday March 04, 2010 @09:53AM (#31357360) Homepage
    The AP he used was a Netgear WNDAP350. There was a typo in the article.
    • by jafo ( 11982 ) *
      Thanks, fixed.
    • A good AP with tons of features.
      Reviewed By: jafo on 2/24/2010
      Rating + 4
      Tech Level Tech Level: high - Ownership: 1 month to 1 year

      Pros: Gigabit port, external antenna jacks, dual simultaneous 802.11n, Linux shell you can login on, Power over Ethernet. Tons of software features: multiple ESSIDs per radio (different authentication can connect you to different VLANs), logging to syslog server

  • by SerpentMage ( 13390 ) on Thursday March 04, 2010 @10:07AM (#31357488)

    Because of the notorious slow conference WiFi's I have learned a new trick...

    I use 3G networks. Since I live in Europe it would be expensive except I get pay-per-day for the country and that averages around 4 to 5 USD per day. That is great considering I can get 3G within restaurants, in my hotel room, and where ever else... Beats having to figure things out with the Wifi...

  • by Enleth ( 947766 ) <> on Thursday March 04, 2010 @10:08AM (#31357496) Homepage

    Even though it's just a short report, it's going to be very valuable for anyone doing similar work, be it for a conference or for a more permanent setup. No textbook is going to protect against those "oh crap, why didn't I think of it before?" moments like some actual experience would, and this posting is the next best thing after actually having someone with experience on site. And this works for any field of applied technology, not just wireless networking.

    So, thanks and be back with some more soon!

    • by Incongruity ( 70416 ) on Thursday March 04, 2010 @10:27AM (#31357676)

      Of course, the top answer to the question you link to comes from Sean (Jafo), the same person who authored the story submitted here. Sean's been nothing short of a hero @ PyCon for a number of years now – the one or two times we tried to replace him with a sub-contracted internet solution, it always ended painfully... or, well, more rightly, with Sean coming in and saving the day.

      So, as someone who has worked with Sean on making PyCon happen, I can say, without a doubt, that he really knows how to get it done. My hat's off to him and

      • Re: (Score:2, Insightful)

        by Anonymous Coward

        If he's so great why did you try to replace him?

        • by AMK ( 3114 ) on Thursday March 04, 2010 @02:06PM (#31360378) Homepage

          For the first year or two, Sean was an unpaid volunteer, and the wireless was an all-volunteer effort. We felt bad about using up all his conference time, so we hired a company to run the network -- they're professionals, so everything should work fine, right? -- and Sean ended up helping them diagnose problems, using up *even more* of his conference time. Now we just pay Sean.

          • Re: (Score:3, Interesting)

            by jafo ( 11982 ) *
            Just to clarify, the only compensation I get for the wireless at PyCon is that my company gets our sponsorship in trade. Usually I even pay for the conference attendance and definitely the hotel and travel. This year was slightly different because the Community Service award I got last year covered the conference attendance and about a quarter of the hotel. PyCon *does* pay for the wireless APs and the like, though I do supply the router from my stash (this year: Atom 330 mini-ITX system, performed admir
  • by mbone ( 558574 ) on Thursday March 04, 2010 @10:27AM (#31357674)

    IETF meetings are larger (1200+ typically), and basically everyone has an uses a laptop / pda, so they make for a demanding wireless environment []. After some really bad experiences, resources were put into this, and the last few years, things have really improved.

    What we have found is that

    - it is necessary to have good gear (not all access points are created equal)
    - To serve a lot of people, lower the power per access point, and put in a lot of them. Raising the power because of poor reception is a mistake.
    - having both 2 GHz and 5 GHz networks really helps.
    - telling attendees how to turn off "ad hoc" mode on their computers really helps.
    - tracking down ill-configured boxes doing bad things on the network really helps.

    Having said that, most recent IETF meeting sponsors have chosen to pay for professional wireless network providers. This is not trivial, and there is no better way to cause a flame war than to have the WLAN melt down.

    • While it certainly doesnt apply to a smaller, less vendor-sponsored conference like pycon, Supercomputing's SCinet [] is always a rather impressive feat. The wireless reception off the main conference floor this past year sucked on 2.4 (I dont know if it was the fault of the convention center's construction plus maybe policies that limited router placement or something), but if you had a device that could do 5 (like most people at the conference), you were golden. Speeds were quite good too, and SCinet handles
    • Is there a good place that lists which equipment is good and which isn't?

    • Re: (Score:3, Interesting)

      by mi ( 197448 )

      IETF meetings are larger (1200+ typically), and basically everyone has an uses a laptop / pda, so they make for a demanding wireless environment. After some really bad experiences, resources were put into this [emphasis mine -mi], and the last few years, things have really improved.

      At what point does it become cheaper (or comparable) to just run a CAT6 cable to every seat in the conference room? I mean, movie theaters and airplanes have that for headphones. Every laptop I've seen has an Ethernet jack... You

      • by Zerth ( 26112 )

        Now if the facility manager thought to do this the next time they redo the carpeting, that'd be awesome for tech conferences.

        But I think having a grid of floor boxes dense enough would make the floor rough to walk on when arranged as a vendor hall or dining room.

      • by dave562 ( 969951 )

        Have you ever actually been to a conference? There are rows upon rows upon rows of chairs in large, or sometimes VERY large rooms. WTF are you talking about, run a CAT6 to every seat?

        • by mi ( 197448 )

          There are rows upon rows upon rows of chairs in large, or sometimes VERY large rooms

          OMG, "rows upon rows"? For a total of — pinky-to-mouth — 1000 people? Puhlease... Sure, it is going to cost money, but so will arranging a WiFi access for the same crowd.

          WTF are you talking about, run a CAT6 to every seat?

          Ever been to Metropolitan Opera? There is small display in every seat showing English translation for the currently-performed aria — and I doubt, MET is unique among opera theaters in that

      • by mbone ( 558574 )

        When you are only meeting once in a given location, get access to the venue a few days in advance, and also don't have the ability to rewire things, putting in jacks and wiring for 1000+ people is not competitive. It's hard enough just giving them a power jack (an IETF requirement).

        Plus, people much prefer the wireless. Even in areas where wireline Ethernet is available, most people use the wireless.

        • by mi ( 197448 )

          Yes, if you don't own the location, you don't want to do this... I had the impression, though, that you do. Or, at least, meet at the same location every time. If I owned a conference hall, I would have done that — and billed a little extra going forward.

          Plus, people much prefer the wireless. Even in areas where wireline Ethernet is available, most people use the wireless.

          Can't explain this... In my opinion, WiFi is the "Plan B" — a fall-back if the real Ethernet is not available... Perhaps, the

    • - To serve a lot of people, lower the power per access point, and put in a lot of them. Raising the power because of poor reception is a mistake.

      What about directional antennas to have any given AP hear fewer clients?

    • by jafo ( 11982 ) *
      This is all sound advice that agrees with my experience. However, this year we really had much less of a problem with Ad-Hoc networks, this was a bigger issue in the past. Possibly some of this is the tools, one of the vendor wireless configurators I saw was basically impossible to tell that it was setting up an Ad-Hoc network, everything I saw in it indicated it was connecting to an existing network not setting up it's own, until I drilled down into the bowels of that software.

      One thing I didn't really m
  • Piffle. (Score:3, Interesting)

    by dtmos ( 447842 ) * on Thursday March 04, 2010 @10:28AM (#31357692)

    The bimonthly IEEE 802.11 standards meetings are co-located with other 802 wireless working groups (802.15, 16, et al.) and regularly have attendance from 600-1000 persons, substantially all of whom are active on 2.4GHz (802.11b/g) substantially all the time the meetings are in session (it's required to register session attendance, upload and download documents, etc., but is largely used for Internet-based multitasking). These networks have worked flawlessly for years. They are specially-built for the meetings by VeriLAN Event Services [], a company specializing in network services for special events. Their web site claims that they have supported events with up to 5000 simultaneous users.

    • Re: (Score:3, Funny)

      by mbone ( 558574 )

      Verilan and Swisscom are the IETF's providers at the present, for when the sponsor doesn't want to do it themselves.

      When sponsors do do it themselves (generally because they sell wireless gear) I would advise them to be afraid. I still remember a poor sales-engineer from a previous meeting (that did not go well in a wireless sense) being told they had implemented some piece of the standard wrong, by engineers who had helped to write the standard. After a few rounds of that, he started visibly flinching when

  • by Andy Dodd ( 701 ) <{atd7} {at} {}> on Thursday March 04, 2010 @10:30AM (#31357708) Homepage

    I don't like the dual-band routers much - they always seem to do a crap job serving both bands, even in the rare cases that the router supports it.

    $300 each for those Netgear APs sounds ridiculous when you can get carrier-grade equipment (such as Ubiqiti Rocket series units) for far less. Instead of getting dual-band stuff, just set up independent 2.4 GHz and 5 GHz networks.

    • by G00F ( 241765 )

      I found it interesting that they used netgear products as well. I use to use a lot of netgear, however, the stuff I have used in the last 5 years have been complete and utter crap. Dieing in a year, and the replacements dieing in a year as well. Although, they do have a nice list of supports xyz technologies.

      I know their office grade is different than their home/soho stuff but my home network which was built 2-3 years ago had 6 different netgear products now only has 1, replacing them with Dlink. In fac

    • GAH. Somehow I typed this all up but forgot to hit submit...

      My experience is that in addition to few routers/APs doing simultaneous dual-band, many don't allow you to do different modes on the two bands.
      802.11n degrades severely in the presence of legacy devices, and it's a spectrum hog. So N in the 2.4 GHz band is a bad idea - my experience is that every 2.4 GHz N solution I've worked with has performed worse than a good 802.11g router with an external antenna.
      802.11n works pretty well in the 5 GHz band

    • by jafo ( 11982 ) *
      The Ubiquity radios were seriously considered, but I definitely wanted MIMO and simultaneous 2.4 and 5.2 GHz in one box.

      Remember, this is all volunteer run, so doubling the number of APs to setup, tear down, and manage was something I wanted to avoid. Also, the Ubiquity gear mostly is not table-top form-factor. It's mean for pole-mounting. Not a good match for our needs.

      As far as doing a crap job of serving both bands, that's been my experience with very cheap units. These units on the other hand, both
  • Test, you idiot (Score:3, Insightful)

    by cerberusss ( 660701 ) on Thursday March 04, 2010 @10:31AM (#31357718) Homepage Journal


    Crimping your own RJ45 should be avoided

    Author should have said "testing should NOT be avoided".

    I hate it when people say such things. A cable tester costs $15 and you neglected testing. Don't say "crimping your own RJ45 should be avoided". That's blaming someone else for your idiocy.

    • Re: (Score:3, Insightful)

      by ZorinLynx ( 31751 )

      As someone who crimps our own cables at work, full agreement here.

      I rarely have problems with self-crimped cables. Another group that crimped its own cables here was having all sorts of problems. Why? They were doing a lousy job of it.

      Test every cable. Make sure the conductor order is correct. Make sure the conductors go all the way into the connector to the stop at the end. USE THE RIGHT CRIMPER. Some cheap crimpers don't crimp all the crimp points and leave the wires less mechanically supported. The crimp

    • by v1 ( 525388 )

      The other thing I was wondering about is the use of stranded or non stranded cat5, and the associated ends you need to use. Last major crimping I had to do, I was provided with solid core cat5 and ends for stranded. (with spades in the ends) You can imagine how that goes, makes for incredibly unreliable crimps trying to use stranded ends on a solid cable. (I don't expect vice-versa to be much better)

      Have they settled on a standard yet? Solid I hope. It's been a few yrs for me.

      • Re: (Score:3, Informative)

        by Kizeh ( 71312 )

        Solid is for horizontal runs (within walls), stranded is for patch cords that will be flexed, such as between a jack and a user's computer. As it's always been.

        • by v1 ( 525388 )

          As it's always been.

          Perhaps. But I have yet to work for a place that stocks both kinds of cables and ends.

          I doubt the flex properties of stranded has much of a net benefit for use in a patch panel... the RJ45 ends are going to wear out long before solid cable. And I've never seen solid get kinked in a patch panel. (contrary to seeing numerous examples of a long run getting kinked)

          • by Kizeh ( 71312 )

            Well, not necessarily in a patch panel, but on the office side where the cable gets moved on a daily basis, kicked around etc. We've certainly seen solid conductors snap after enough flexing. (Then again, when you roll chairs and filing cabinets over patch cords, they all get demolished sooner or later ;-) )

            My employer did stock both, when we still did patch cables in house.

    • by TheLink ( 130905 )
      In most cases it's better to buy pre-crimped certified cables and test them, than to crimp them and test them.

      How much money are you going to save? If you're charging $$$ for the project, crimping your own cables is a waste of time and resources. Unless of course your job is making cables :).
      • Except in temporary setups where you don't have specified lengths or you have lots of wildly different lengths. By the time you measure out a run and test a pre-crimped cable you could have already installed the uncrimped cable. This also doesn't mention the fact that pulling cables with ends already on them is quite difficult compared to just straight cat5 or cat6 especially if you're using conduit.

        We used to do the whole pre-crimped thing, costs us a lot less money and we can deploy a lot faster just mak

    • A cable tester costs $15 and you neglected testing.

      A $15 cable tester will tell you 1/10th as much as you could find by looking at the crimped ends of the cable. You'll get the wonderful green light on the cheapo tester no matter how much attenuation, crosstalk, etc, you're getting with a junk cable.

      To really know you've got a halfway decent cable, you need a cable tester that runs around $1,000USD. Do a search for "certification tester" and try to find a $15 unit. And they're just getting more expensive

    • by jafo ( 11982 ) *
      I appreciate the feedback, but, I stand by my original statement. Crimping RJ45 ends on cable should have been avoided. We wasted easily a couple of hours of scarce volunteer time on this. Yes, a $15 tester could have been used, and almost certainly wouldn't have detected either of the primary problems we had with the crimping.

      However, the primary reason to avoid the crimping is not a technical one, it's a management one. We really didn't have the resources to be spending on the crimping. We should hav
  • ...So you pretty much learned:
    • Not to be a novice and get your enterprise and common-sense hat on for a big event?
    • How to a proper WIFI site survery prior to deployment?
    • Not to make your own hacked-together setup (This isn't your mom's basement, buy/use legitimate, reputable and trusted equipment, cabling, software, ect.
    • Using SOHO/home networking equipment for THAT many potential users?

    Learning is good and you were successful for the most part. Regardless of the downplay of comments you'll receive here on

    • Re: (Score:3, Insightful)

      Actually knowing a bit about jafo and the setup at PyCon, I can tell you exactly what's wrong with your idea: money.

      The IEEE/ACM SuperComputing trade show's network (SCinet) does exactly what you say. They also have 10,000 attendees, over 50 people working on the network, a decent budget, and a ton of donated gear and bandwidth.

      PyCon (and jafo) don't have $100k to spend on the network. That means that you have to make do with low-cost commodity hardware. The fact that the network can stay up and deliver acc

  • Hmmm.... that's odd.... the Cylons must have managed to just miss all of Picon's wireless access points with their barrage of nukes. Or maybe it was just part of their "plan" all along?
  • one of the guys couldn't establish a PPTP connection

    PPTP? People should know better. PPTP is very [] weak []. OpenVPN and IPSec are much better all around.

  • This year I tried to get public IPs again, but we just weren't able to get any meaningful number of them. So, we had to do NAT. This worked well, and I had no complaints beyond the first one: one of the guys couldn't establish a PPTP connection. I had forgotten to load the NAT protocol modules... I loaded those up and it went smoothly after that.

    And you think that's going to be more likely to happen in future years?!

    Next time, set up IPv6. Use 6to4 tunneling if you can't get an actual IPv6 drop from the network provider. But just get IPv6. Then everybody will have a public IP.

    • Except that more than likely the vast majority of attendees won't have IPv6 enabled on their machines or are running XP or older. IPv6 in XP is crap especially if you're trying to do a 6to4 setup.

      The problem with conferences is that you have to accommodate a wide range of equipment. I'd love to just deploy 802.11n at my events except that I've yet to see any vendors or attendees with equipment that supports it. Hell, for a lot of them 802.11b is still the only choice. So I'm left with needing to expand 802

    • The network was set up with low-end gear and a lot of ingenuity to save money. I don't have a whole lot of experience here, but I haven't had a low-end router yet that understood IPv6. I'd really like that to change, but I think it's going to be a few years.

      • Only the router(s) need to understand IPv6. I would imagine that could be handled with one router running Linux, possibly with a trunk port to a managed switch that did 802.1q. Not that expensive, and they likely already have that anyway.

  • * My laptop only does 2.4ghz n, and I thought that was par for the course? Are laptops with 5ghz N really that common?
    * There is clearly a quality difference among access points, but how do you tell in advance which equipment will work and which won't?

    • by AK Marc ( 707885 )
      My laptop only does 2.4ghz n, and I thought that was par for the course? Are laptops with 5ghz N really that common?

      For the past 5 years, I've only bought 5 GHz wireless cards. Sure, most of them can also do 2.4, but 5 GHz works so much better because it is comparatively empty. I've never had any trouble getting 5 GHz for a laptop, and n came out for 2.5 and 5 at the same time.
    • by mbone ( 558574 )

      I believe that every Mac laptop since 2006 [] has supported 802.11a. Joining an "a" network is seamless, and much of the time you won't even notice it.

    • by jafo ( 11982 ) *
      See the graphs in the article, but the majority of users were indeed running 5.2GHz and 802.11n.

      I tell by getting candidate APs and testing them. The Netgear I used had issues when running WPA, but as the event was running wide open, and my testing had no issues, I was happy.

      The gear I really wanted was 3Com's equivalent, for 33% more, but my vendor said they didn't expect to get the first shipment until the week before the conference. Not enough time to test. And I'm glad I didn't even think about it --
  • This year's FOSDEM [] in Bruxelles had over 2400 unique MAC addresses and 3600 visitors a day(source []). We enjoyed a 1Gbps pipe, and far from saturated [] it.

    It was overall of excellent quality, though there was a glitch in at least one of the hacker rooms where the operators had to upgrade the AP firmware. The geographic setup was more broken out: FOSDEM happens at the Universite Libre de Belgique (how appropriate), with talks in lots of classrooms spread across a few buildings.

    It would be useful for everyone if

  • I wouldn't exactly call PyCon's Wi-Fi a success, but it was better than in years past. The venue changes every two years, so all the bugs have to get worked out in each new location then it's better the second year. I mostly used my iPhone on 3G and stayed off of Wi-Fi. The Hyatt must have had a micro-cell in the bldg because I had a strong signal 2 floors below street level.

    If you tried to use wi-fi during the keynote session in the morning, it was slow as molasses. During the regular conference session
  • Some questions:
    • Why do you need a few minutes to talk about the network? It seems to me that 30 seconds spent listing the network names, passwords, and wired port locations is more then enough. Assuming it's an open network; the 600 people who used it didn't need to talk about it. Perhaps some "network posters" would be more helpful as they can be referenced throughout the convention
    • What's wrong with NAT? It's the norm when using an open network; and it gives you (and the attendees) an ounce of protecti

Machines that have broken down will work perfectly when the repairman arrives.