News for hackers. Stuff that matters.
Re:CouchDB ships 1.2.0, Windows packages, new website by rogerbinns on Saturday April 07, 2012 @01:03AM (#3810021)
The one thing CouchDB does really well is multi-master replication. Any server can be in any state and replicate with any others. It works somewhat analogously to a flood fill. Contrast with other databases where you have to nominate master servers and communicate with them.

I gave up on CouchDB two years ago. Partly it was a poor experience with the community, but mostly technical issues. Writing is done to append only journals. This means not only the added/changed data is written but also all changed btree nodes. I was starting with around 2GB of JSON data (10 million documents). With CouchDB it would turn into 43GB on disk (including views/indices). With tuning, tweaks and compaction I was able to get it to half of that, but it was still a massive explosion of data size.

I tried MongoDB with the same data and ended up with just over 4GB consumed on disk. In addition my data load time was 6 minutes compared to 75 minutes for CouchDB. I was even experiencing unindexed queries in MongoDB being faster than indexed ones in CouchDB!

I'm sure CouchDB has improved a bit since then, but I don't think the implementation can do much better especially at the data sizes I was doing. (Personally I think that is a rather small amount of data, one of the CouchDB developers called a million documents large!)

There are no comments on this comment yet.