Eduardo Gargiulo wrote:
Nate Duehr <nate@natetech.com> said:Eduardo Gargiulo wrote:Hi all, this is my first message to this list. I would like to know if there is a formula or method to calculate the minimal bandwidth requirements for a given network. I'm thinking to pay a data link to feed a debian box and from there, via wireless links (802.11b), give internet access to aprox. 15 users. I want to guarantee 128 kb/s for every node. Which is the best way to calculate this? Any hint would be appreciated. Thanks in advance, and sorry for my english (i'm from Argentina). --ejgYou've already got it in your original message.15 * 128KB seems to be a relatively simple calculation. Plus overhead if the server is actually doing server duty like e-mail or web.Not sure what the question is here.I think there was a missunderstanding of my problem (probably because of my poor english). I don't think that the relationship between bandwidth and simultaneous users using that bandwidth is linear (FIXME). The problem is one internet connection to a linux box, and 15 wireless users (simultaneously connected in the worse case) with 128Kbps guaranteed. The question: How much bandwidth should I purchase to satisfy that requirement? You mean 15 * 128Kbps = ~2Mbps ??? I think it's too much for 15 users!! regards, --ejg
I understand.Basically the question is "unanswerable" unless you know the useage patterns of those 15 users. Best case, only one user is online and can use the entire pipe. Worst-case, all 15 users are online doing something that requires a continuous stream of data at 128Kb/s (VoIP, downloads, etc...).
Almost all ISP's oversubscribe their upstream links, but you used the magic word "Guarantee" in your message, and that changed my response.
If that guarantee is a business contract obligation to your customers, you should build for the worst-case scenario, or close to it, especially if the contract includes refunds for non-performance.
(I used to work at a data center company that truly had a money-back guarantee if the customer couldn't run their full data rate to where-ever they wanted to go on the Net, but of course -- we couldn't guarantee that the far-end could handle that. Downtime of routers to backbone providers in that environment could lead to massive cash refunds if the problem was created by us.)
If by "guarantee" you mean you'd *like* to offer them at least 128Kb/s per user under "normal" conditions -- but you have no requirement to do so -- then you can get away with buying less upstream bandwidth and monitor useage carefully.
The business model can be done a number of ways. High-end customers *will* pay for that sort of guarantee. Residential/casual Internet users with no need for that level of service typically won't.
You'll probably find that there will be 2 or 3 users out of 15 that really use bandwidth heavily, all day long, and the rest will be in bursts, depending on when they're awake, what type of network applications and useage they're doing, etc. The "average" web user will use bandwidth to check e-mail, go to websites during peak (evening) times, and maybe an occasional download (Windows Update, grab the latest Linux ISO, send a bunch of photos to Aunt Millie) on a regular basis.
You can also determine what type of bandwidth useage is appropriate for each price point by setting the rules and building a way to monitor bandwidth use on a per-IP basis.
A user who wishes to run a bittorrent server with tons of popular (let's assume they're all legal for this discussion) files might suck down bandwidth continually all day long.
Figuring out the bandwidth needed for such "average" users to remain happy with the service while not realizing that your upstream bandwidth is at maximum capacity is an interesting exercise, written about in many whitepapers over the years and also partially an art form.
The key is MONITORING. You must build an infrastructure where you KNOW which users are using up the pipes.
Example: My ISP is fairly liberal in their policies. I have an overall monthly bandwidth "cap" limit that if I go over it, I start paying extra for the priviledge of using that much bandwidth in a month on my 1.5 Mb/s DSL line. I calculated the cap one time and found that I'd have to run 21 days at full data rate on the DSL to hit the cap... something I'll probably never do. However, recently I've been enjoying listening to mp3/podcasts from various sources that update daily, and my usage has gone up dramatically, even considering there are servers here.
They also allow me to up my data rate (faster DSL circuit) with or without upgrading my cap. The pricing for each is separate. So if I feel like downloading things faster and am willing to run the risk that my faster line can get me in trouble sooner on the overall bandwidth numbers, I can assume that risk by simply ordering a higher speed line.
Upstream bandwidth is similarly "capped" but since my DSL is assymetric, I will never hit the cap during a month on the outbound side. It's not possible.
Hope that helps, Nate