Confluence Data Center testing gets slower with more nodes available

Hi there!

We have used the AWS templates to build a Confluence Data Center environment. We have started running some tests in a large instance and we have realised that each test gets slower with more nodes.
For example, for a view page test we get:

  • 1 Node: 280ms
  • 2 Nodes: 301ms
  • 4 Nodes: 390ms

Does this happened to someone else? What could cause this scenario to happen? We understand that the numbers should be the other way round, the more nodes you have the less response time you get…

Many thanks in advance!

1 Like

Do you have your app installed? Is there something causing node communications?

Hi Daniel,

Thanks for the quick reply. The tests have been done without the app installed and there is no communication issue between nodes.

What else could we check?

Hi Patricia,

We recently went from 6 Confluence Data Center nodes to 3 and have seen great improvements in response time. By reducing the number of nodes, there is less Data Center communication that takes place thus freeing up CPU and memory cycles for your actual cluster to perform.

In addition, we increased our nodes from an M class to a C class with additional memory and have seen significant CPU Performance improvement using our new nodes.

You might find this article helpful when sizing your cluster. This gave us the insights we needed to change our nodes to give us the performance we needed. Infrastructure recommendations for enterprise-scale Confluence instances

I hope you find this useful.
Joe

1 Like

When running E3 (which also uses the Atlassian Cloudformation AWS templates) against a vanilla Confluence this is what we get:


This would confirm your assumptions.

But then there is also this picture in the official Atlassian docs which makes it look like performance going down with more nodes seems to be expected…

dc-scale-measures

Now I’m confused as well.

1 Like

Hi,

Those Atlassian results make now more sense, as is exactly what is happening to us :woman_facepalming:

So now I’m a bit confused and don’t know what to expect :woman_shrugging:

Joseph, we also increased the machines to C5 and we are running confluence with 12GB of memory, which we think should be enough… but still having the same results :frowning:

We were only able to improve the results by replicating the indexes properly between nodes

Thanks all for your replies!

1 Like

Hi Patricia,

By upgrading the servers to C5, did you also remove one or two of the servers? Having the same number of servers is not likely to improve your performance due to the same amount of Data Center traffic.

Another resource for you to consider is Tips for building an effective performance monitoring strategy. Also, my team lead spoke at Summit 2019 on the The Top 5 Skills Enterprise Admins Need to Know

Reviewing these resources should provide you with the information you need to tune your Confluence Data Center for optimum performance.

If you have additional questions, please let me know.

Joe

1 Like