NoSQL data stores

Here is the list I composed during the weekend:

• Redis – No big name (yet?)
• MemcacheDB – Used by sina.com.cn
• MongoDB – SourceForge
• Tokyo Cabinet and Tokyo Tyrant – Used by mixi.jp
• Cassandra – Used by Facebook and Digg
• Voldemort – Used by linkedin
• Riak – No known web sites yet.
• CouchDB – Used by BBC and Ubuntu (the Ubuntu One project)
• Dynomite – A clone of Amazon’s Dynamo, seems no user yet.

I’m not sure if I should try out these projects, the major concern is that I don’t have confidence on grid (hadoop, etc) about their latency:

• Hypertable
• HBase

and I won’t consider these because they are relying on specific providers:

• BigTable
• Dynamo
• [Could be more ...]

Java and erlang are pretty popular here – I don’t like one of them, and I don’t know much about the other, tough.

2 Comments

• 

  By Hang, 2009-10-27 @ 12:47

  Seems this blog is doing better than I did, and I should do some evaluation on these:

  • Flare
  • LightCloud, it is based on Tokyo Tyrant
  • KiokuDB
  • Persevere
  • CloudKit
  • Lux IO
• 

  By Hang, 2009-10-28 @ 09:49

  I think I got some idea about what’s my requirement:

  • Surely performance is one of the key factor, but as long as it’s much faster than MySQL (I’m talking about 2X~10X) then it is acceptable
  • Query feature, key/value pair is good enough, but if it has further support on (kind of) complicated queries without performance lost, then it becomes a plus
  • Reliable persistent storage. It does not have to consistent for all the time, but eventually it should be. Also, reliable means it won’t lose writes (rejection write is ok).
  • Built-in replication is a must, and if it is easy to setup/monitor, then it will be a better choice, if it has automatic failover feature, then it becomes great.
  • Sharding, better on server. If it is in client side, then language supported will be a key factor – I know I’m going to need C/C++, Perl, PHP, Python, and Java
  • Language used to develop server. This is less important, but if it is done by some language I’m familiar with, then a plus, right now I’m seeing C/C++, Java, and Erlang, and my preference is exactly the above.
  • Misc features, just nice to have – authentication, statistics, packaging (RPM and deb), any big names are using it, etc.

  I will quickly rule out some projects mentioned here and then focus on only candidates with good potential.

Other Links to this Post

RSS feed for comments on this post. TrackBack URI

Leave a comment