| |
|
Tuesday, August 27, 2002
|
|
Google Rank
Today (or sometime before today, but now's the first time I realized it) I moved to #1 in Google for "Brian St. Pierre". Woo hoo!
(Ignore the fact that numbers 2 through 20 or so are that other Brian St. Pierre.)
6:56:22 PM 
|
|
Communities of Scale (2)
Beyond the infrastructure, you have to think about how the content of all those weblogs scales. Sure, you can put 10-20k weblogs on a single static server and fit the 5-10 required servers in a rack. (Note: I don't know a whole lot about the resources required for this; I'm trusting John Robb's numbers here.) But how can you find what you're looking for over 10-20k weblogs?
It's very easy to throw hardware and bandwidth at sites and make them scale. The costs are more or less distributed depending on your architecture. What's difficult is building a scalable community, of finding like-minded souls. And thus we have the Radio Community Server, the Blogging Ecosystem, blogdex and others — or the less-sophisticated GeoCites Member Pages directory.
[The Peanut Gallery]
This point can't be emphasized enough. Without some kind of way of either organizing or being able to search (or filter) the information on a large number of weblogs, you end up with the chaos that is the Web, albeit on a smaller scale (but without Google!). So it is crucial to be able to organize those weblogs into some kind of useful structure.
I'm watching the action on the ecosystem/indexing front to see what happens. (And participating on the blogchalk front, although the organizational scheme and the idea behind it is much less formal or useful compared to, say, the Blogging Ecosystem.
8:20:06 AM
|
|
Communities of Scale
The Peanut Gallery discusses a John Robb post on "Communities of Scale":
If done centrally, you could probably put a thousand or two weblogs on a single server. That would take 50-100 servers, extensive rack space, and a huge budget for admin of those servers given that there is complex functionality on the server. In a decentralized model, you could put 10-20 k weblogs on a single static server. That would require only 5-10 servers (a single rack) and a very low admin budget.
This depends entirely on how the system's built.
First, let's distinguish between the centralized and decentralized models. In the centralized model, all functions are hosted on the server-side. In the decentralized model given here, the server exposes your works to the world, but editing and other tasks are completed on your desk. The classic web hosting arrangement is this decentralized model. For ease of use, to get away from the complexity of synchronizing filesystems over FTP, many vendors provided web-based HTML editors, and moved to the centralized model. But at some point it becomes easier to buy GoLive or DreamWeaver — unless you're using Blogger, Moveable Type, or one of the Userland tools. Think of the server as your upstream cache.
In a centralized environment, the systems performing the editing tasks are often segregated from those serving the final product. Usage shows that pages are read more frequently than they are written, so the editing systems need only support a small fraction of the total membership. If you store the data separate from the presentation, and render the two on the fly, then you run into some processing bottlenecks, but most on-line editors generate static HTML of one variety or another. The point being that you don't need 50-100 servers for 100,000 weblogs unless you're Manila.
[The Peanut Gallery]
Using the Internet as a prime example, I think it's much easier to scale a decentralized system. You mostly have to add resources at the edge, where necessary. The impact of growth on those few centrally located resources is less than in a centralized system. Especially if you're talking about the (relatively) small amounts of bandwidth and storage required by webloggers.
One issue with the decentralized model is that security can be difficult to implement -- generally speaking, it is easier to secure something when all the resources are concentrated. In many enterprises, this is something that can't be ignored.
8:07:54 AM
|
|
|
© Copyright 2002 Brian St. Pierre.
Last update: 9/6/2002; 11:12:15 AM.
|
|
|
