How do you build a scalable infrastucture for a family of related but separate websites?

So, my company is talking about building an ecommerce platform that will serve many different clients. Each client would have a different look and feel, and its own set of users, but the backing code (ie: adminstration services, authentication servers, checkout services, possibly admin pages, etc.), and some users would be shared, so a bug fix could be applied to all sites at the same time, a primary admin could log into all websites.

As the entire StackExchange set of websites (with pretty high traffic) run off a small number of servers (two I believe) I wonder what it would involve to serve up many unrelated (but similar) websites through one webapp, or even one database.

In order to have one database, I imagine that every table would have columns identifying which realm the entity belonged to, and every SQL call would filter by that column. That seems like it would become an maintenance nightmare, and (less importantly for me) a DBA's hell.

Another option, with one webapp, but multiple databases, I imagine the realm could be tied to a specific data source, where all non-shared data could be specified. Then when any request was made, the appropriate datasource could be loaded and the webapp would run as if there was only single source. This would have an 开发者_如何学JAVAadded benefit of being easily horizontally scalable since the exact same webapp, but a different set of realms and datasources, could be spawned when necessary. Websites could be easily moved to new servers as well by simply copying the webapp and moving the database.

I wondering what other possibilities there are, as well as, specific examples if they're out there.

Note: I'm not talking about Twitter-scale scalability, nor about hardware/languages/etc, but rather design methodologies and patterns.

The architecture you are talking about is called "multi-tenant" architecture. There are diff. ways of going about architecting a multi-tenant application. Broadly speaking, data tier can be architected in 3 ways:

1. seperate database for each client - easier to code, diff. to maintain
2. one database, different schema
3. one database, one schema (with clientid in each table; except for meta-data tables) - more time in coding, easier to maintain

Each has its own advantages/ disadvantages. Take a look at this article by Microsoft on multi-tenancy. http://msdn.microsoft.com/en-us/library/aa479086.aspx

Broadly speaking, I would suggest option 3 as it offers true multi-tenancy. If you have some tables that you expect to become very large, you can partition that table based on the clientid (for eg: if you want 10 partitons, you could do partition based on mod of clientid)