Please see http://www.nosqlbenchmarking.com for more informations and benchmark results

You can now build this project using Maven. As a few dependencies where not available in any maven repository, I have decided to provide them myself in the lib directory. You still have to add those jar to your local maven repository, to do so execute the script addToLocalRepo.sh or execute each of the three lines that it contains. For Cassandra and HBase, there is a little more configuration needed. For Cassandra edit the file storage-conf.xml to reflect your cluster configuration. For HBase edit the file hbase-site.xml to reflect your cluster configuration.

Then you only have to run “mvn package” to build the jar that contains everything needed. It will be in the folder “target” and it will be called “Wikipedia-nosql-benchmark-1.0-SNAPSHOT.jar”.

You can use the jar as a single client or with a controller and several clients if you need (they are mainly usefull if the client’s bandwidth is too small). You can also use this jar to insert the documents in the DBs and to measure elasticity.

Call the jar file like this :

java -jar jarfile.jar fillDB dbType /path/to/documents/ numberOfDocuments nodeAdress firstID numberOfInsertRun

dbType can currently take the following values :

• cassandra
• scalaris (please note that the current implementation should be optimized)
• voldemort (no MapReduce implementation so only read/update can be benchmarked)
• terrastore (partial implementation, needs work)
• riak
• mongodb
• hbase

/path/to/documents/ is the path to a directory that contains files named as increasing integers going from 1 to numberOfDocuments
nodeAdress is the IP address of a working node in the cluster that can handle writes
firstID is the integer that should be used as the starting ID of the inserts, this is usefull is you already have inserted documents and want to increase the data set with the same documents but using new IDs
numberOfInsertRun is the number of time you want to insert all the documents

To benchmark the read and update performances call the jarfile like this :

java -jar jarfile.jar runBenchmark dbType totalNumberOp readPercentage numberOfDocuments IP1 IP2 IP3 …

dbType can currently take the following values :

• cassandra
• scalaris (please note that the current implementation should be optimized)
• voldemort (no MapReduce implementation so only read/update can be benchmarked)
• terrastore (partial implementation, needs work)
• riak
• mongodb
• hbase

totalNumberOp is the total number of operations that will be requested to the cluster.
readPercentage is the proportion of requests that will be read-only
numberOfDocuments the total number of documents already inserted in the DB
IP1 IP2 IP3 … is a list of IP that can be as long as desired. For the systems without a load balancer, I put in the list the IP of each
active node in the cluster. For systems with a load balancer, I simply put the IP of the load balancer as many time as there are active
nodes in the cluster.

If you want to benchmark the MapReduce performances , call the jar file like this :

java -jar jarfile.jar dbType search numberOfRuns IP
dbType can take any value listed just before provided that this database has a MapReduce implementation.
search is a constant
numberOfRuns is the number of time that the search benchmark will run
IP is the IP of one of the nodes of the cluster that accept MapReduce jobs

Please note that the elasticity test is only available in controller/client mode

This mode use configuration files that must be in the same directory than the jar :

• benchmark_clients must contain the IP addresses of all the servers running a client, one IP per line without any blank line
• benchmark_nodes must contain the IP addresses of all the servers that should be benchmarked
• elasticityArgs must contain the following line if you want to use the elasticity test :
  – the upper limit acceptable for the standard deviation on 10 runs (use a statistical test at 95% for example)
  – the lower limit acceptable for the standard deviation on 10 runs
  – the number of milliseconds that must be waited for between runs
  – the maximum number of runs

First you have to start the clients on each server that you specified in benchmark_clients, to do so simply run the command
java -jar jarfile.jar client on each of them.

To benchmark the read and update performances call the jarfile like this :
java -jar jarfile.jar controller dbType totalNumberOp readPercentage numberOfDocuments

Please note that the arguments are exactly the same than for the standalone client, except for the IP list that is now provided in the file benchmark_nodes

If you want to benchmark the MapReduce performances , call the jar file like this :
java -jar wikipedia.jar controller dbType search numberOfRuns
The arguments are exactly the same than the ones for the standalone search benchmark, except for the server address that will be used. Note that the IP that will be used is the first one in the file benchmark_nodes

If you want to start the elasticity test that will run until it sees 5 consecutive runs whose standard deviations are within the bounds defined by the two first lines of the file elasticityArgs or until it reach the maximum number of run defined in the same file. Call the jarfile like this :
java -jar jarfile.jar controller dbType totalNumberOp readPercentage numberOfDocuments elasticity

Finally you can kill all the clients from the controller by running the command :
java -jar jarfile.jar controller kill