CouchDB vs. MongoDB Benchmark

Edit (9/1/10): this benchmark is old, silly, and should probably be ignored in favor of more recent and representative ones. I don’t want to take it down for historical purposes, but seriously people, it was never a good benchmark, it’s over a year old at this point, and both databases have changed a lot.

Edit (12/6/09): this is the #1 Google result for “mongodb benchmark”, so I figure I’ll do some community service: if you’re interested in benchmarks, you might want to look at the 3rd party ones listed on the mongodb.org website.


Felix Geisendörfer did a benchmark in PHP that was super-easy for me to port into MongoDB. You can see his post on his blog.

And now… comparing his results for CouchDB with mine for MongoDB’s (I did the graph in Open Office, which is why the quality sucks):

As you can see, MongoDB does, uh, slightly better.  Here are the numbers:

# of Inserts Couch Total Time (sec) Couch Time/Doc (ms) Mongo Total Time (sec) Mongo Time/Doc (ms)
1 .0015 1.46 .0005 .5
2 .0015 .75 .0004 .2096
3 .0017 .56 .0005 .1604
4 .0017 .44 .0005 .1190
5 .0018 .36 .0005 .1060
6 .0019 .32 .0006 .0931
7 .0021 .3 .0006 .0847
8 .0022 .27 .0007 .0789
9 .0023 .25 .0007 .0734
10 .0025 .25 .0007 .0721
50 .0072 .14 .0024 .0476
100 .0136 .14 .0044 .0442
500 .0687 .14 .0253 .0505
1000 .1361 .14 .0372 .0372
2500 .4686 .19 .0278 .0111
5000 .9165 .18 .0488 .0098
7500 1.5116 .2 .0835 .0111
10000 2.3111 .23 .1065 .0107
25000 6.8684 .27 .2711 .0108
50000 15.8227 .32 .5430 .0109
100000 35.3071 .35 .1.7697 .0177
250000 104.0009 .42 6.4533 .0258
500000 230.6021 .46 11.7684 .0235
750000 352.7959 .47 17.0473 .0227
1000000 487.3284 .49 18.4376 .0184

Please let me know if I made any mistakes, all the values were hand-copied.

I ran these tests using the PHP driver on Ubuntu 9.04 on my MacBook Pro.  You can see the test script I forked on Github.

A little analysis: Both DBs start with some overhead, but by 1000 inserts CouchDB seems to be chugging along nicely.  MongoDB takes slightly longer to hit its groove, hitting its peak around 10000.  They both slow a little near the end, as MongoDB starts spending most of its time allocating files and, although I know almost nothing about CouchDB’s structure, I’d guess it’s doing something similar.

  • Jan
  • Jan
  • http://www.snailinaturtleneck.com/ Kristina

    I would second the vote for getting kv store benchmarks out there.

    And, as you requested, MongoDB data out to 10 million records:

    http://snailinaturtleneck.com/downloads/couch.vs.mongo.csv

  • http://www.snailinaturtleneck.com Kristina

    I would second the vote for getting kv store benchmarks out there.

    And, as you requested, MongoDB data out to 10 million records:

    http://snailinaturtleneck.com/downloads/couch.vs.mongo.csv

  • patsplat

    CouchDB's performance is terrible. Query performance doesn't get much better.

    In recent benchmarks for a client, Mongo is the fastest, Cassandra is slower but presumably some of the overhead is the p2p autosharding, and CouchDB is absurdly slow compared to the two leaders. Futon looks pretty tho.

  • Ted

    this is an absolutely absurd comparison. you are talking about running two very different datastores on a *single* node (a MacBook Pro at that), with a bunch of other non real-world assumptions. show me the following performance benchmark: replicated mongo on 10 nodes vs couch on 10 nodes. then show me performance when the network goes down or there is a disk failure. (see my point?)

  • kristina1

    Of course it's absurd, but I didn't write it. I converted a CouchDB benchmark someone did because I hate writing benchmarks. I used a MacBook Pro because that's what I had.

    You propose an interesting benchmark. If you implement it in CouchDB, I'll implement it in MongoDB :) I don't know much about CouchDB's failover, it might outperform Mongo's (but I doubt it). Only one way to find out!

  • Aaa

    2500 inserts, MONGO TIME/DOC (MS) should be 0.0111.

  • Guest

    I like this kind of smartness on you .. thanks, you made my day …

  • kristina1

    I'm not sure what you mean, can you clarify?

  • aha

    couchdb is super slow and it crashes very often…

  • Jason Goemaat

    27.8 ms total time / 2500 inserts = 0.01112 ms per insert

  • Anonymous

    Ah, thanks! Fixed.

  • Frank

    Where can I find the test scripts?

    http://github.com/beaddy1238/couchdb-benchmarks/tree/master

    gives a 404 :-(

  • Anonymous

    I changed my Github username since, so it’s actually http://github.com/kchodorow/couchdb-benchmarks/tree/master.  I’ll update the link.

  • Frank

    Thanks a lot

  • Markus Fenske

    Yep, you made a mistake. “.1.7697″

kristina chodorow's blog