Confluence has been updated to version 6.15.9

This article describes the performance of the Exalate app.

The performance can be defined in many different ways with multiple measurements. 

It also depends on different factors such as

  • size of the data being exchanged
  • number of simultaneous sessions
  • available resources on the issue tracker ( memory, CPU, disk speed...)
  • versions and type of the issue tracker


We've performed the transactional performance test under the load to answer the following questions:


  • How long does it take to process 1 synchronization transaction when there is no load on the system?
  • How long does it take to process 1 synchronization transaction when there are hundreds to be processed?


Environment

Simple synchronization setup between Jira Server instances.  

Outcome

Exalate is already in use since 2014 at different enterprises. The Exalate has been configured as a single-threaded application which processes the synchronization transactions one at a time.  This has 2 consequences 

  1. A minimal system load, as the exalate can be compared to a single user who is working hard to update/retrieve information
  2. It takes time to process transactions (because of the single-threaded nature) and the product is not meant to do large migrations

One deployment has on average 12000 synchronizations/month without a significant impact on the overall instance performance


Test results from 2017

1. Bulk create test

  • JIRA Server to JIRA Server (Both Software)
  • Generated 10 000 issues with 69 585 comments (using the JIRA Data Generated)
  • Synchronized issues so that they've been created on the remote side
  • This took 38 hours - or an average of 14 seconds per issue

2. Bulk comment test

Added 1000 comments to the set of issues under sync

The following steps have been performed

  • Setup
  • Pause the synchronization
  • Bulk Edit (using the standard JIRA functionality) and add a small comment
  • Resume the synchronization
  • This took around 400 seconds to process, which is 2 comments per second

There was no noticeable load on the application itself (both JIRA's running on the same machine which mimics a bit of load due to other activities)

The average load of the machine never exceeded 0,5.