By all means, start out doing table joins inside the DB. I have relatively rarely seen a situation develop where the schema vs. the needed data (and how often and how quickly) ends up showing how you've painted yourself into a bit of a corner and you can work around the problem by doing some joins in the client.

Much of the time, this is also about limitations or quirks of the query optimizer and how much time / expertise you can throw at recrafting the query to convince the optimizer to do the efficient thing.

Frankly, I've more often seen client-side joins done because the people working on the project rather sucked at database performance issues. Or even because the team just believed that "joins are expensive". But when I've actually worked on the performance issue that lead to client-side joins, there was a much more efficient solution that didn't involve client-side joins.

So, I have done client-side joins for performance reasons. But only very rarely and always due to things that had better resolutions in the long run. It can be a powerful work-around.

I've never seen client-side joins done at the outset for good reasons. So, yes, start out with server-side joins, by all means.

Joins are costly, and considering there typically are more clients than servers, there are a lot of situations were it makes sense to do the join on the client -- there's a lot more client CPU available than server CPU.

And if a database join seems "expensive" to you, I suggest you need to look at your indexes.

Try another rule of thumb: the network is slow. Moving more data from server to client than you need to is rarely a good idea. First guess-- albeit not the last word-- is to use the features built into the database to crunch the data.

I will repeat myself, but I fear I might as well talk to the walls: it all depends. Measure, and you will know. Don't assume the network is slow -- and that hence you should overload your database server as much as possible. Slowing everyone else down, so your query gets done "at the server" may not the best solution.

Measure. Don't assume.

Measure. Don't assume.

And once you have implemented a solution because it's the best today, measure the next day again and the day after, and keep repeating it, because circumstances change.

That's right, you've got to measure everything, everything! When I design a system I always write the code every possible way and then benchmark each of them against each other before roll-out.