DB benchmarks: Cassandra vs. BigTable vs. Hadoop(s)

Question

I am looking to evaluate the possibility of using Cassandra, BigTable, or a Hadoop-solution. Are there any places that have an up-to-date comparison on how these three compare and perform on a set of benchmark tests? I found a few from perhaps five years ago, but I would like to get a better high-level view of the three databases -- their strengths and weaknesses -- before diving in.

Answer 1

1. Hadoop is an ecosystem, having many types of databases running on it. What i believe you want to compare is HBase. It is slower that Cassandra in most benchmarks i've seen, you can easily find benchmarks for comparing these two. And when you are talking about Hadoop you need to understand that this is much more complicated environment than Cassandra or BigTable.
2. BigTable can eat pretty much all you throw on it, just pay google and all will be ok. (Seen benchmark with 2 million record/second write).
3. Cassandra architecture is based on DynamoDB(AWS) and BigTable design. Is very fast in workloads it is designed for (you can find many benchmarks for 1 million writes a second).
4. Never trust benchmarks on internet :). You must run it by yourself, with your sample data. Not all databases suits all needs. If you provide a bit more information about what workload you want to put on databases, people here could guide you for the database/databases you need. (there are a lot more than the 3 you've listed).

Answer 2

YCSB (Yahoo! Cloud Serving Benchmark) is the standard for performance testing NoSQL databases. This presentation seems to sum up the goal of YCSB of testing latency and scalability.

In addition to latency and scalability, I check throughput (ie total work done) for high batch load, and tail latency for "online" type systems when the servers are adequately provisioned. By tail latency I mean things: how fast is slowest 1% (or .1%) of requests under.

Cloud Bigtable put out some YCSB results when it first launched (FYI I work on that product).