Scalable web-based applications has got a lot of air-play on social networks like Twitter lately, mostly because Twitter has been overcome by scalability issues and the service unreliable. Having an unreliable internet service is embarrassing and unacceptable. There is no better way to kill your brand image than to have your viral social media strategy fall flat on its face in front of the “instant on” world.
Having said that, I’ve had a lot of requests for guidance on how to build scalable web-based applications that can withstand getting Slashdotted. Building a scalable web-based application can happen incrementally so you don’t have to apply all the principles below all at once. There is an appropriate implementation roadmap that is appropriate for different types of web-based applications. For example, if your site is proving an web-service API (SOAP service) for other web-services or client applications, you should spend time looking at capacity planning for those services with less emphasis on your own public facing web-site. Anyway… here is my brain dump. There is more where this came from.
- Judicious use of the singleton pattern
- Judicious Use of the Concurrency pattern
- Client side form validation
- Use of AJAX for web-services requests
- Caching of semi-dynamic data (pre-rendering semi-static pages)
- Use of MVC pattern
- Object persistence is separated from the object
- Use of XML and meta-data instead of traditional row/column SQL commands for each data element
- Place SOAP interfaces on a separate cluster of servers (aka application servers)
- Use a load-balancer and multiple Web servers
- Separate your web-application into a minimum of four tiers: UI, Application Services, Object Persistence Layer, Database Services
- Use a separate NIC on each server and switch for each tier of the application (e.g. web-page server to application server, application server to object persistence server, and object persistence server to database server)
- Use a load balancer between each tier of of the application (e.g. between web-page servers and application servers, between application servers and object persistence servers, etc.)
- Use an enterprise service bus object persistence service to make sure object concurrency issues are handled across multiple databases
- Use clusters of virtualized servers running across multiple physical servers
- Add performance monitoring services on each virtualized server to check memory, network, harddrive and application utilization
- Conduct performance tests to determine if any one object needs more server resources, and move those objects onto their own virtualized and physical servers
- Use lots of partitioning of data across NICS and Physical Switches (aka switch processors)
- Use firewalls in front of each switch in the top three tiers (UI, Application, and Object Persistence Layers)
- Place network probes between each firewall and load-balancer to monitor utilization and intrusion detection
- Place each application stack, a complete web application, application server and database server, in at least two data centers in two different geographic locations using two different internet data service providers
- Create a private point-to-point network between data centers for transaction load balancing using two different internet date service providers
I know was a lot of technical mumbo-jumbo. Frankly, I haven’t met that many web designers, web-developers, or even a lot of software developers that understand everything I’ve listed. In order to implement the list, it will take more than one skillset from several technology professionals. (Business owners can contact me privately about assessing the maturity of your development staff. You might be surprised.)
If you don’t understand it, I am happy to elaborate privately. Understand that this is what I do when I’m not playing drums in a Jazz combo.
The information contained on this post is my opinion, and mine alone (with the occasional voice of friend). It does not represent the opinions of any clients or employers.