How can I determine the amount of bandwidth needed to support the following scenario?
- 1,000 simultaneous WiFi users all on same network
- Web surfing
- Downloading applications
|
How can I determine the amount of bandwidth needed to support the following scenario?
| ||||
|
feedback
|
This question came from our site for computer enthusiasts and power users.
|
You might already know most of this, but since the question was originally asked on SuperUser I want to cover the entire scope of the problem. You see, you really have two problems: the wifi network and the bandwidth question. BandwidthI'll deal with your bandwidth question first. I run the campus network for a small college with 450 students plus associated staff and faculty. At this size, we get by (barely) with a 20Mbit (up and down) direct fiber connection from Time Warner. We're hoping to increase this in the near future, but 20Mbit is good enough that our students can do things like watch streaming Netflix movies at reasonable quality most of the time. However, this works largely because of our gateway software (we use untangle) and time I put in monitoring and adjusting for traffic patterns: using QoS to control things like Netflix, windows updates, apple downloads, etc, and using other options to come down pretty hard on prolific p2p users. Since it looks like you're roughly twice our size, that means you'll want at least a 40Mbit connection, and since the overhead administering the network tends to increase at a greater rate you'll probably want to do more like 80Mbit, as that's likely cheaper than paying a dedicated employee to monitor your bandwidth use full time. This is especially true if this is a short-term event rather than a long-term installation, as part of what makes tuning the lower bandwidth possible is sticking with it long enough to recognize your user's trends and needs. Another rule of thumb is to take a typical low-end residential dsl connection (say 768Kb down, 128Kb up), oversubscribe it 6:1, and allow that much bandwidth per user. However, this doesn't always scale well, as the more users you have the more likely it is that most of them won't be online all at the same time, and you end up paying for bandwidth you don't need. WiFiNow let's talk about that 1000-user wifi network. There's a lot to cover here, so I'll take it in steps: CostAgain, this depends a bit on whether you're doing a one-time event or a long-term deployment. For a one time event in a large open meeting space, you can probably get by with about 20 users per access point, with the access points spread evenly throughout the space. Turn the transmit power down, carefully balance the channels by hand, have some good monitoring software going, and that's about as good as you can do. All told, that will require at least 50 access points. The going rate for enterprise-grade equipment is around $600. That yields a total of $30,000 — before you even worry about any controllers, gateway, installation, or software expenses. The good news is that for an event like I'm talking about you might be able to rent much of what you need. For a long-term deployment it gets worse. You're spreading out over a much larger area and there will be more environmental and structural factors (walls, stairwells, bathrooms, etc) to account for. You need to think in terms both of basic coverage for some areas and density for others. Expect to at least double the total number of access points you want (probably triple), because at any given moment you'll need the same amount of capacity as you did for the large room, but your users won't be spread out as evenly and much of your equipment will be sitting idle. You'll also need multiple controllers at $2000-$5000 a pop. VLANS/Broadcast domainsOne important thing to remember is that cells in a wifi network are (to put it in wired ethernet terms) half-duplex, unswitched, and highly sensitive to packet collisions. In a nutshell, that means you really need to worry about the size of your broadcast domains. I'm fortunate, in that our network is just small enough that I've been able to get by with a single vlan for wifi, but I know that for best performance most larger enterprises look to start splitting up their wifi into multiple vlans with chunks of no more than around 500 clients each (or even just 250). This is enforced by dividing your deployment up into different zones or areas (you might, for example, have one zone per building), each with a different network address range enforced via subnet masks set at /23 or /24, and sometimes an exceptionally short dhcp lease time (I've seen as little as 5 minutes, whereas the default in Windows Server is 8 days). It can make roaming from one end of campus to the other on a single SSID difficult (if that's a design goal), but some vendors have controllers that support this. Density/Interference issuesAnother problem with wifi is RF (Radio Frequency) interference. A given access point can only support so many clients at a time with the available spectrum. It's tempting to just add access points, but you have to be careful because as you do this you have not increased the amount of spectrum available and at the same time you have increased the amount of RF noise in that spectrum. In effect, just adding access points can often make things worse. It's also worth noting that the biggest source of interference for a given access point is very often a neighboring access point from your own network. A common strategy to solve this problem for dense areas in a wifi deployment is to greatly reduce the transmit of power of each access point, so that each one is responsible for a much smaller space (cell size). This allows you to install the additional access points you need without just creating additional interference. An added benefit is that a good controller will be able to detect access point failures and automatically increase the transmit power of neighbors to compensate without any noticeable outage for your users. Unfortunately, low transmit power is only a partial solution for the density problem, as you don't have control of the transmit power setting on your client radios; enough individual clients transmitting loudly in a small area and you'll be SOL. This is not yet a fully solved problem, though I understand 802.11n does a lot better. ChannelsYou also want to be careful which channels you use. Wifi signals (like other RF signals) tend to bleed over quite a bit into neighboring channels. For example, in the U.S. there are 11 channels allocated for 802.11g wifi, but really only three of them should be used in most circumstances: 1, 6, and 11. A good controller will take care of this for you, but absent that you'll need to carefully balance your deployment to avoid putting too many of the same channel access points too near each other. Don't forget to think 3-dimensionally, if you have access points on multiple floors in the same building. Also don't forget that the signal broadcast by a client at the edge of one cell can be heard by access points outside of that cell, so when allocated channels a good rule of thumb is to allow double the typical radius of your cell size between channels whenever possible. SecurityThe final piece of the puzzle here is security. You probably don't want to run an open network, but you do want to make it easy for users and guests to get online without too much fuss. The current solution is called 802.1x, often using a RADIUS server. It works pretty well, but this is also the single largest source of issues posted to some of the wireless mailing lists I follow. In other words, it's not the 100% solved problem your vendor may have you believe (though it's getting close). | |||||||||
feedback
|
|
Quick answer: figure out how much bandwidth you want each user to have, add in an absolute minimum of 25% for overhead, lost packets, RF problems, etc. and multiply by number of users. We give each user 2Mbps, so for 1,000 users, that's about 2.5Gbps total. I don't know where you're located, but that's about the level where I'd consider getting colo space in a peering facility and buying commodity Internet. Not-so-short answer: As you can tell from the previous answers, this is a VERY complex question. I'm part of the team that runs, among other things, the multi-campus wireless network for a major state university. At this moment, we have 5,315 users logged in via wireless, from around 1,000 access points. Many of those users are idle, or are phones or other devices that associate whenever they're within range. We're about saturating 1 Gbps of uplink at the moment. The network I just described is running on smart controller-based APs that cost about $1,000 each, $500,000 of controllers, and has a full-time staff of four with over 30 years combined experience. And we ONLY run the wireless portion, we don't own a single switch or router or fiber line. I don't know what type of infrastructure you have in place, but for 1,000 users on wireless, your needs will be far beyond "buy 1,000 linksys routers and plug them in". You'll need to think about doing RF surveys of the locations where the APs will go, channel plans, etc. There will be a lot of bugs to work out. And you'll need an enterprise-grade physical infrastructure (routers, switches, etc.) as well. If you're starting from scratch, I'd highly recommend bringing in some contract help to assist with the initial design, equipment selection, etc. and training your staff. BTW, I'm one of the four full-time staff, I spend about 25% of my time doing actual wireless work (helping others), and the rest dedicated to back-end systems (DHCP, Radius, Syslog, DNS, etc.) and tool development. | |||||||||||||
feedback
|
|
Good articles on successful setting WiFi at the PyCon conference (from 600 pax in 2007 to 1000 attendees in 2010) in years 2007, 2008, 2010. It covers all the aspects - number of APs, frequencies, protocols, trends, lessons learned, ... - including info on bandwidth consumption. | ||||
|
feedback
|
|
How long is a piece of string? If you are talking about streaming websites where everyone will be on at the same time, you will most likely need a hefty leased line connection or similar. If you are talking about a few news/optimized websites and you have a proxy/caching server, you can get away with a standard ADSL connection. You really have not given enough information and it is far too hard to help you based on that. | |||
|
feedback
|
|
First of all, you're probably going to need multiple wireless routers because the speed does degrade as more users log in (even when they're not doing anything). The reason large companies with hundreds of computers (let alone a thousand) need to use ethernet switches rather than hubs is that they need to isolate background network chatter (e.g., ARP traffic) and reduce collisions (this wasn't a problem with Token Ring due to the way the topology was designed, but there were also disadvantages with Token Ring that ethernet is immune to). If you put 1,000 computers on a single wireless network, you're most likely going to experience some serious congestion problems as more and more users get connected, even if they're just sitting there doing very little with their internet access. How much web surfing and downloading do you anticipate? Will they also be using eMail and VPNs, Skype, and other applications that use different protocols? Be sure to factor in DNS, DHCP, and specific web sites like YouTube and SlashDot. My recommendation is to figure out first what the usage will more likely be (it helps to know who these 1,000 people are), and try to work from there. I think you're probably going to find that you'll need to have multiple wireless access points unless you can find a very heavy-duty one that is designed to deal with this type of load. | |||
|
feedback
|
|
1,000 similutaneous WiFi users all on same network Good luck with that! You will need a lot of access points. Overall, for that amount of users you will need to do some simulations to determine the likely bandwidth needs. | |||
|
feedback
|