This is a fork + visualization of: http://wiki.github.com/brianfrankcooper/YCSB/

1. Download the latest release of YCSB Visualization version:

   curl -O --location https://github.com/ArchangelX360/ycsb-visualization/releases/download/0.10.0-visualisation/ycsb-0.10.0-visualization-release.tar.gz
   tar ycsb-0.10.0-visualization-release.tar.gz
   cd ycsb-0.10.0-visualization-release

2. Set up a database to benchmark. There is a README file under each binding directory.

3. Run YCSB command (See YCSB visualisation for more info for the visualization).

   On Linux:

   bin/ycsb.sh load basic -P workloads/workloada
   bin/ycsb.sh run basic -P workloads/workloada

   On Windows:

   bin/ycsb.bat load basic -P workloads\workloada
   bin/ycsb.bat run basic -P workloads\workloada

Running the ycsb command without any argument will print the usage.

See https://github.com/brianfrankcooper/YCSB/wiki/Running-a-Workload for a detailed documentation on how to run a workload.

See https://github.com/brianfrankcooper/YCSB/wiki/Core-Properties for the list of available workload properties.

YCSB requires the use of Maven 3; if you use Maven 2, you may see errors such as these.

To build the full distribution, with all database bindings:

mvn clean package

To build a single database binding:

mvn -pl com.yahoo.ycsb:mongodb-binding -am clean package

This YCSB module extension stores YCSB measurements in real-time into a MongoDB database. It is designed to be used with Web Dataset Visualizer out of the box.

Place in the overall project architecture

This module provides a new thread that periodically checks if there is new measurements and if any stores them into a storage DB. We created a new MeasurementType "frontend" which is using a concurrent map of lists of MongoDB Documents instead of basic(operation, latency) couples. Choosing this MeasurementType will automatically launch the whole DB storage process. The MongoDB connexion will be initialized based on specified parameters or the default local ones and the periodic fetching process will be started.

We added some parameters to YCSB:

• frontend.collection.benchmark=<my_bench_name> where "my_bench_name" is the name of the benchmark's collection in the storage database. Be careful, if you use this name for two different benchmarks, the result of the second will be considered as new results for the first one.
• frontend.collection.counters=<my_counters_collection_name> where "my_counters_collection_name" is the name of the counters collection of the storage database. One you chose one, try not to change it because you will have to reconfigure your client depending on which benchmark you want to watch.
• frontend.db.uri=<my_db_uri> where "my_db_uri" is the MongoDB URI without the database name.
• frontend.db.name=<my_db_name> where "my_db_name" is the name of the MongoDB Database where to store benchmark results.
• frontend.frequency=<my_frequency> where "my_frequency" is the number of milliseconds between two storage actions in MongoDB. (5000 by default)

All the above parameters have default values thus are not mandatory.

As we said above, we also created another type of measurement:

• measurementtype=frontend will activate the frontend DB hook for measurement and allows the graph visualisation

A local benchmarked database is drastically reducing performance, we recommand you not to use it. We chose not to compare results with local benchmarked database as it is a rare YCSB user case.

Using medians of 20 iterations in the range from 15k points benchmark to 1M points benchmark we are :

• never under 6% slower for the both LOAD and RUN phase
• never under 1% slower for the both LOAD and RUN phase with remote storage (most common use case).

Using minimum values of 20 iterations in the range from 30k points benchmark to 1M points benchmark we are :

• never under 10% slower for the LOAD phase
• never under 12% slower for the RUN phase.

Below these values, results have a strong variation and no conclusion can be made. This is not a problem for us as our extension module provide no added value for small benchmarks. Indeed, they are too fast to take advantage of a real-time display.

Sometimes our module is faster on Throughputs but has a larger execution time due to databases connexions and storage thread initialization and shutdown. We invite you to consult the report for more information.

The storage process, as it is implemented, should work regardless of the database thus you can use your custom adapters. However, it has been tested with memcached only.