Ryan Tomayko rebuked Daniel Sabbah of IBM’s claim where LAMP cannot scale and the model should “grow up”. He argued the traditional 3/n-tier design cannot scale due to its complexity, and a simple state-less design would be more scalable, faster and easier to develop. I am not really convinced, and I do hope that they do have some really really really fast object serialisation to keep very complicated data objects/relations always in the DB (instead of partially cached in app. server). And what if your fully-transactional DB is the bottleneck?
Scott Yang's Playground
Faith, Technology and Randomness in Life, According to Scott
Jun 8 2005
The problem with 3/n-tier application deployments and scalability relates directly to the RAS (Reliability, Availability and Serviceability) infrastructure built-in to the application, or lack there-of. When problems occur and you don’t have the tools to trace a transaction thru all the tiers (and back), then the app. can’t scale. Or worse, it will act fine for awhile, blow-up, then days of downtime to find and hopefully fix or get a fix for the problem.
Communicating these problems is exceptionally difficult as most folks think of scalability in terms of tps, not can you fix it if it breaks. I’ve told one client their 100M dollar app. won’t scale; their response is that they are ahead on the hardware curve. When D-day happens, it’ll be a sad day.