Nick Granado - Likes

While the memory based part is good, Jim Starkey seems to fall in the dead category as well in several places. The following two quotes seem to stand out: "Web scale databases like BigTable and SimpleDB make simple things difficult." and "Developers are used to SQL and are comfortable working within the transaction model, so the transition to cloud computing would be that much easier."... more... - Knut

5 more comments

I'm a developer and like SQL. I waist a lot of time trying to manipulate the OO layer, when many times writing my own queries would be much easier and more efficient. - Chip Ramsey

I'm not saying SQL and RDMS is always wrong. I have done my share of SQL and as you say, I always ends up writing SQL myself instead of using OO layers. I developed a tagging server (http://sourceforge.net/project...) using SQL directly. When you need more than two way relations, key-value stores aren't as well suited. I am currently using CouchDB for a project and the API and map-reduce views are sweet. - Knut

Knut those were my quotes and come from experience developing on those systems, following forums, and talking with developers. Many things in apps are simple gets/puts of structured objects. But many times you also want to know simple things like sums, averages, counts, top 10, etc while not having to do all the heavy lifting in application code. SQL is not the only option. MS has DryadLINQ, for example. But the point is scaling can't be complexity shifted to developers if adoption is to be wide. - Todd Hoff

If you look at CouchDB, you will see that you can do counts etc with map-reduce. You can write them very easily in javascript and you will have access through the same API you use to access key-value objects. Needless to say, you can to the same with map reduce on bigtable as well. As for scaling, I still don't agree. RDMS system are hard to scale and companies have traditionally hired database admins to tune them. I don't get how that should be harder with key-value stores. My experience is the opposite. - Knut

On who's cluster are you running the map-reduce jobs Knut? Interestingly enough map-reduce isn't part of Google App Engine. And this is more of declarative-compiled-into-map-reduce-etc is good versus hand-coding-the-same-thing-all-the-time is bad argument than RDBMs are great if you just give them a chance argument. - Todd Hoff

CouchDB uses map-reduce internally for views. You write the mappers and reducers (if needed) in javascript. In other words, you don't actually run the map-reduce yourself, you add in a similar way to a stored procedure and you can actually do this in Futon, the CouchDB admin interface. Which cluster you run this on doesn't really matter much as long as it can run CouchDB. Google App... more... - Knut

Nick Granado › Likes