JIRA performance on Linux

[Ronny]

Hi all,

I understand from reading some threads and also web article that JIRA seems to be running faster on unix machine than windows.

However recently I did a comparison of doing a system restore, one is on a windows app server with a postgreSQL on windows, and the other system is Linux Sles 10 app server with PostgreSQL which run on windows os.

The restore only took like 2.5 hours on the Windows system and 5 hours+ on Linux system.

Below are their cpu and memory spec for the app server (please note that both systems are virtual machine):

Windows :

2 cpu + 4 GB ram

Linux

8 cpu + 32 GB ram

There must be clearly something wrong here, perhaps my JVM settings ??

I set the windows one with 2GB JVM heap size and 8 GB on the Linux.

Im confused now, did I just make the wrong decision in hosting JIRA on linux box ?

[Hemant]

Both Linux and Windows are adequately supported by JIRA and JIRA do not necessarily support one environment over another. Performance wise there wouldn't be much difference when both environments are configured properly. This is especially because JIRA runs in a platform-independent java environment, so as long as the server has the required hardware resources (RAM, CPU, Excellent disk speed etc) JIRA should perform similarly in both environment.

Perhaps something to consider would be the general idea that Linux systems often require less RAM to run than windows. This may in some scenarios (especially if you have limited resources) imply that you get more efficient performance from Linux compared to windows. However, this is not necessarily a JIRA or java application specific issue.

For more general references on Supported platforms and requirements please see the following documentation:

• https://confluence.atlassian.com/display/AdminJIRAServer072/Supported+platforms
• https://confluence.atlassian.com/adminjiraserver071/jira-applications-installation-requirements-802592164.html
• https://confluence.atlassian.com/enterprise/jira-sizing-guide-461504623.html

Thanks,

Hemant

[Nic Brough -Adaptavist-]

No, it's a good decision, but I suspect you've got the right idea when looking at the memory settings.  With java systems, "more memory" is not always a good way to improve things.  It's far more complex than this, but I tend to think a simple way to explain it is "Too large a stack can mean the system getting bogged down in looking after it instead of running the application".   As a general rule you can only really usefully compare performance of a java application running on Windows with the same application running on Linux when they have identical memory and cpu settings (then you'll see you can get up to 25% faster response and update times from the Linux box!)

It's four years old now, but Dan's talk on it at Summit a few years ago is still well worth watching - http://www.adaptavist.com/w/atlassian-performance-tuning/ (Yes, I work for Adaptavist, so I am sneaking in a bit of an advert, but the talk is not a sales pitch, and I did help a bit by being the audience for six rehearsals of it)