Ask HN: Why do web frameworks always use database?
3 comments
Taking 'information heavy' to mean large data size/ volume. File system can be seen as a basic database. When data access paths are not pre-defined e.g. URL, slug, tag, etc., a systematic way to organize and query is required. Three examples: (a) not knowing what user may query, text indexing today almost index every words. (b) given predefined access path, NNTP have used mostly files. (c) mail servers have both. IMO, in general, addressing domain specific requirements, database systems (relational, logical, network...) are not slower, harder to update/ backup. As pointed out in another comment, disk IO is slow. Both file and database systems feature various ways to speed that. Understanding requirements, then architect, design, and choose a solution stack carefully.
I'm designing an information-heavy web site (static content, no user content), and have been wondering the best way to do it.
Look into Movable Type (http://movabletype.com/). It is written in Perl and it writes static html files. I did the redesign of http://www.steves-digicams.com/ in Movable Type 4.23 and there are over 12,000 pages. Movable Type generates the static files at publish time in about as much time as Wordpress takes.
A big advantage of using static files on Steve's Digicams is that it is very easy to replicate across a load-balanced network of servers without the hassle of having to load-balance MySQL with a master-slave configuration.
Look into Movable Type (http://movabletype.com/). It is written in Perl and it writes static html files. I did the redesign of http://www.steves-digicams.com/ in Movable Type 4.23 and there are over 12,000 pages. Movable Type generates the static files at publish time in about as much time as Wordpress takes.
A big advantage of using static files on Steve's Digicams is that it is very easy to replicate across a load-balanced network of servers without the hassle of having to load-balance MySQL with a master-slave configuration.
Thanks for pointing me to MT. I looked at it a bit, and it seems promising. It looks like it stores everything in a database, and generates the static files from that. How do you deal with a major site update? Would you copy the database to the dev machine, do the update, and then have some SQL magic to merge in only the parts that have changed? (Since you can't copy it back, as then you'd lose all the forum posts since the copy-to-dev)
Disk I/O is a huge limiting factor in a filesystem based approach to data persistence. Using a database for the persistence of data allows the db to schedule flushing the state to disk, independent of the actual requests. The situation you're specifically talking about, where there's nothing but static pages, there's no point in using a framework that's designed around building dynamic pages. You could just as easily write/use a templating system that takes content files and generates the HTML files from that content, and just run it when you have new content.
Wouldn't disk I/O be equally bad for a database? I assume the database gets cached in memory (as much as possible), but I would think that the file cache would have the same effect.
The database can make all the disk I/O operations asynchronous because it provides an interface on top of file system itself. What you had listed would require either your own implementation (which you'd be better off using existing DB/Berkeley DB systems) of similar functionality, or actual disk I/O.
* All those database lookups are going to be slow
* All your data is locked up in a binary file
* How do you update your site? It doesn't seem like you can easily merge two databases, or update just the change deltas.
Instead, if all the content lives in separate files with other files telling how to put it together, then you can use a compiler (of sorts) to generate the site as a bunch of static pages for the site. The advantages would seem to be:
* Fast to serve content
* Easy to redesign (edit files, re-run `make` on the static pages)
* Easy to update the server: cvs update -d -P; make
* If there is user content, CGI could add files (assuming availability of locking). Then a backup would be as simple as `cvs commit`.
* Rolling back after an update gone bad would be easy.
The drawbacks seem to be minimal:
* Requires extra disk space (cheap)
* CPU-intensive to change site layout (infrequent)
* CPU-intensive to add content (presumably rarer than reading content, which is fast)
So am I missing something? Why are databases so popular, given that they seem to be slower, and harder to update/backup?
[Edit: slightly better formatting]