Wikibench Performance on c3 Instances

Today, we explore the performance of wikibench on ‘c3’ instances. The methodology we used is similar to the one we utilized on ‘m3’ instances in this post here.

In testing ‘c3’ instances, we used a paravirtual image for all our experiments. We used Ubuntu 12.04 LTS (64-bit architecture) for our OS. Another major difference in this set up compared to the one for ‘m3’ instances, is that we used the SSD instance drive for MySQL DB storage. If you recall, we had used EBS store for ‘m3’ instances

In logging our performance comparisons, we have used the following metrics

Missed Deadlines

Due to the delayed response time from the server for lower sampling rates, we see that wikibench is not able to replay all the requests in the tracefile. This results in missed deadlines. Here is a table of missed deadlines for each of the instance types.

table 1


chart 1


Request Timeouts

We did not record any request time outs, except for ‘c3.large’ when the request rate was 46239 requests per hour. This is a significant improvement when compared to that of ‘m3’ instances.

Average Response Times

As we did for ‘m3’ instances, we calculated an average response time by first calculating the averages over a period of one minute and then calculating the average of the resulting values.

Here is a table of average response times.

table 2

Following is a chart for the inflection points for each of the instance types. We see that the inflection points for ‘c3.2xlarge,’ ‘c3.4xlarge’ and ‘c3.8xlarge’ are same.


chart 3


CPU Usage

This table shows a breakdown of CPU usage averaged over the total number of CPUs in each instance type. The average is taken during the ~20 hour period, which includes requests replayed from all the trace samples.

table 3

chart 4


Memory Usage

Here is the table that represents the average memory usage during the ~20 hour period, which includes requests replayed from all the trace samples.

table 4


chart 5


Clearly, we can see that on ‘c3.4xlarge’ and ‘c3.8xlarge’ not all the memory is occupied. This also corresponds with the fact that we see very little difference in performance between ‘c3.4xlarge’ and ‘c3.8xlarge.’

Let’s keep the conversation going about wikibench performances on instances. Inquire with us at Or, learn more about Flux7 and its solutions, even find out about our assessment packages, at

June 10, 2014 / Benchmarking

About the Author

Flux7 Labs
Find me on: