Couchbase Google Compute Benchmarking

Welcome!

This repo contains scripts that will allow you to do the following:

• Create Couchbase Server and Couchbase Client instances in GCE
• Initialise the Couchbase cluster
• Install our benchmarking load-generator on a number of clients
• Invoke the load generator and record some stats
• Post-process these stats ready for analysis

Pre-requisites

These scripts assume you already have access to GCE, have gcloud installed and are authenticated to a particular project. The scripts source the "settings" file for all the configuration of GCE and Couchbase. If you want to use the Couchbase UI externally, you'll also want your project's network settings to allow access to port 8091.

Creating your instances

First you'll want to create some instances. This is performed by the create_instances script.

There are 4 stages to this:

1. Create the Server template ./create_instances.sh create-server-template
2. Create the Client template ./create_instances.sh create-client-template
3. Create Server instances based on the template ./create_instances.sh create-server-instances

• Instances are created in parallel-batches so the output can get garbled.

4. Create Client instances based on the template ./create_instances.sh create-client-instances

The Server and Client templates are customised by the setup_cb_image, setup_cb_instance and setup_client_image scripts respectively.

Initialising the Couchbase Cluster

This is done by clusterInit script and there are two stages.

1. Cluster init
2. Bucket creation

For both steps you will need the external address of one of the couchbase nodes and the couchbase-cli in your PATH.

Cluster Init

The parameters are external-address and cluster-quota-MB.

# 50GB cluster
./clusterInit.sh create-cluster 104.154.61.11 50000

Bucket Create

This step can be repeated to add more buckets.

The parameters are external-address, bucket-name and bucket-quota-MB

1GB bucket called charlie
./clusterInit.sh create-bucket 104.154.61.11 charlie 1000

Installing the Load Generator

Next you want to install our load generator (libcouchbase's "pillowfight") on each of the client instances. This is done by build_install_pillowfight.

Generating Some Load

Now it's time to run some load - do this by running run_pillowfight_3B. Again note the variables at the top that dictate how many clients, how many documents in the dataset etc. Log on to the Couchbase UI (<cb-server-node-external-IP: 8091

Capturing the output

The load generator selects a machine and runs the cb_perf_monitor script on it. This creates a series of .out files on that Server node's /tmp directory. These should be copied locally for further post-processing and analysis. The first client instance also produces a "durability_perf." file which captures the latency timings for the operations.

Post-processing the stats

stats_report.py takes the .out files generated above and converts them to several SVG files, using pygal.