-
Suggestion
-
Resolution: Fixed
-
None
-
Confluence is running in Amazon EC2 as outlined in Adrian's blog here http://blogs.atlassian.com/developer/2009/04/setting_up_jira_and_confluence.html.
We evaluated Confluence 2.10.3 on an Amazon EC2 instance as outlined in Adrian's blog here http://blogs.atlassian.com/developer/2009/04/setting_up_jira_and_confluence.html. We've noticed that Confluence running in this configuration can be sluggish when first loading pages. We're currently using Trac hosted on a machine in our office, which is pretty snappy. I signed up to the Confluence 3.0 hosted trial to see if the response times were any better, and it seems to be about the same as our instance running in EC2. We have a good net connection here at the office (ADSL2+), so I don't expect that's a factor.
YSlow grades Confluence an E when navigating to a space for the first time. Notice that 28 Javascript files are loaded. These really should be aggregated in to one javascript file for performance.
See the attached pictures for the YSlow and Firebug reports.
Any suggestions on how to get Confluence running better in this configuration?
- relates to
-
- Closed
-
- Closed
[CONFSERVER-16115] Confluence is extremely slow to load pages when network latency is significant
Jol Blazey
added a comment - Firefox Net chart for baseline server root page, compared with confulence wiki page from same network
Jol Blazey
added a comment - Firefox Net chart for baseline server root page, compared with confulence wiki page from same network Jol Blazey
added a comment - I checked the network latency of the server by running ffox Net, on our server root tomcat page. There are 4 requests for small gifs and html, each of about 300ms which can be regarded as a baseline ping. (see screen dump ec2 latency)
Requestin a confluence wiki page on the same server, (within 5 min of the baseline) took about 16 sec and looked like there were approx 30 + individual GET requests ranging from 200 - 500ms. (see screen dump 30 plus http get) Some of these will be in parallel but that's still lots of heavy lifting that could so easily be optimised.
If we moved our hosting to AU i doubt whether we would improve our baseline much more than 300ms. I really think it is a too many uncompressed http GETs' issue not server latency.
There is a maven plugin for http://developer.yahoo.com/yui/compressor/ that works well from experience. It would fix this problem.
<plugin>
<groupId>net.sf.alchim</groupId>
<artifactId>yuicompressor-maven-plugin</artifactId>
<version>0.7.1</version>
I would appreciate any work on this as there have been lots of staff comments on the performance of our new wiki, which is otherwise very popular.
Thank you, Jol.
Jol Blazey
added a comment - I checked the network latency of the server by running ffox Net, on our server root tomcat page. There are 4 requests for small gifs and html, each of about 300ms which can be regarded as a baseline ping. (see screen dump ec2 latency)
Requestin a confluence wiki page on the same server, (within 5 min of the baseline) took about 16 sec and looked like there were approx 30 + individual GET requests ranging from 200 - 500ms. (see screen dump 30 plus http get) Some of these will be in parallel but that's still lots of heavy lifting that could so easily be optimised.
If we moved our hosting to AU i doubt whether we would improve our baseline much more than 300ms. I really think it is a too many uncompressed http GETs' issue not server latency.
There is a maven plugin for http://developer.yahoo.com/yui/compressor/ that works well from experience. It would fix this problem.
<plugin>
<groupId>net.sf.alchim</groupId>
<artifactId>yuicompressor-maven-plugin</artifactId>
<version>0.7.1</version>
I would appreciate any work on this as there have been lots of staff comments on the performance of our new wiki, which is otherwise very popular.
Thank you, Jol. 
Adrian Hempel [Atlassian]
added a comment - Hi Sean,
The problem is a combination of the large number of files loaded when you first visit Confluence, combined with the network latency between yourself and your EC2 instance (which is located in the US).
Given that the issue is not specific to EC2 (and given that our support doesn't extend to hosting within EC2), I've updated the title of the issue.
Regards,
Adrian
Adrian Hempel [Atlassian]
added a comment - Hi Sean,
The problem is a combination of the large number of files loaded when you first visit Confluence, combined with the network latency between yourself and your EC2 instance (which is located in the US).
Given that the issue is not specific to EC2 (and given that our support doesn't extend to hosting within EC2), I've updated the title of the issue.
Regards,
Adrian 
We've improved our browser footprint significantly in recent releases of Confluence by reducing the number and size of requests made by Confluence for CSS and JS resources.
We'll continue to work on this in the future, and have further enhancements planned for the next major release: Confluence 4.0.