Is it a good idea to develop a website that has little access to database?

It will be as slow as molasses in February and won't make it more secure. Unless there is a concrete reason for doing so, you will never get benefits from increasing your pipeline length. Every step in your pipeline length has a latency cost of approximately 1 to 10 milliseconds. This results in a huge design restriction in terms of speed and flexibility. So the key question is why? I could see something resulting from a very specific set of infrastructure requirements but outside a very narrow set of circumstances, I cannot easily see a benefit.

Over the years I have worked on a huge number of web applications with many different architectures. Fashions in software architecture come and go in cycles, just like anything else, and we all have a tendency to reach for the tools we know as the solution for any problem.   My analysis of your problem, given the information you have provided, is that you want to reduce the possibility of duplicate logic while decoupling your system components. Both of these are sound architectural principles that should bear fruit with improved reliability and maintainability over the lifecycle of your application. The decision you need to make is the decoupling point.   There is a practical difference between decoupling and remoting. Firstly, it is not necessary to have a remote interface in order to have a common API that decouples business logic from client applications. Secondly, the cost of remoting is latency and latency kills performance (if we are looking at performance from the point of view of the end user). Each remoted operation takes at least an order of magnitude longer to dispatch than a local operation   How visible are the effects of latency introduced by multiple remote requests? Actually this can be easily demonstrated without introducing any application logic. In my framework Fakoli (http://www.fakoli.org/) we typically need to serve somewhere of the order of 100 javascript and CSS files to the browser when we first deliver a web page. Fakoli can put individual script and link tags into the HTML sent to the client, whereby the browser has to make 100 or so requests back to the server, or we can use an optimization feature that combines files of the same type together into larger files. The effect of the optimizer is significant - a non-optimized page can take around 10 seconds to load from scratch, whereas the optimized version can load from scratch in less than a second.     Network latency is a cost that consistently gets underestimated. It was the Achilles heel in Enterprise Java Beans, and it is the reason that we see so many spinners as modern websites load themselves up a piece at a time via AJAX requests. I have seen many EJB applications that tried to get around the remoting performance bottleneck by building 'fat' interfaces to return all the information in one big gulp, which leads to a degenerate data model and potentially a lot of unneeded data being passed across the network.   All this is not to say that remote requests are inherently bad - just that the cost of using them needs to be considered if you are concerned about performance.   I believe one of the key insights here is that while your web-site is conceptuallya client of your business logic, it is of necessity going to be running on a web-server, will be accessed via HTTP/HTTPS, and serves text back to the client as a result of requests. Therefore it has all the attributes it needs to also be your API server. The approach I would usually use in this case would be to build a common API as a library or component, and then have that library expose a RESTful API for remote access to the business logic by clients other than web browsers.   Other concerns I have heard include "our front end developers only know JS/CSS/HTML5 and our backend developers only work in .NET/Java/whathaveyou so we need a way for them to communicate" or even worse "our backend developers don't DO front end design. That is for the designers".  While these are legitimate issues to confront in your development team, these are not architectural issues, but rather political ones.