Can anyone explain the principle of web-based pattern search?

When using a pattern search application (Kombilo or its Windows variant Drago), at initialisation stage, the program process a collection of sgf files, extract pattern information to save in files. I suppose these files get loaded into the memory so search within the application can be quite efficient.

On a web-based application, I imagine one cannot load the pattern files when responding to a user request because loading big files would take too long - wait more than 1 or 2 seconds is probably not acceptable for a web application. So what is the suitable technology to implement web-based pattern search?

Your server has to keep the pattern tables in RAM between requests. You must not start the program from scratch each time. Technology such as fastcgi allows to do this easily.

Thanks for the tip. I am now able to do this efficiently by writing a server loading the pattern tables and then listening to a port. Then a client program send query to that port. Invent a simple protocol and do a little bit socket programming.