Distributed server

[fisher]

The documentation explains how the server can be spread across multiple machines. Has anybody tried this and did it go as planned?

Steve Fisher

[jmoore]

Hi Steve,

The documentation explains how the server can be spread across multiple machines. Has anybody tried this and did it go as planned?

The only "spreading" that we (Glencoe) use in production is the starting of a secondary Java (session) server while another is running so that new code can be deployed with zero downtime. The major requirement on the Java processes (Blitz-*, Indexer-*, PixelData-*), currently, is that they must all have access to a shared file system, preferably NOT NSF.

Beyond this, the following processes can all be put elsewhere requiring only configuration changes:

• Processor-* - Jobs sent to the server will be spread across all available Processors.
• DropBox & MonitorServer - These will need a directory like /OMERO where data can be dropped in (if you're using that)
• The PostgreSQL database itself can run on a separate server.

This can all be done by defining more:

Code: Select all

<node name="whatever">
  ...


instances in your etc/grid/default.xml file. Once a icegridnode process is started on the other machines, it will connect to the main server and download its own configuration. You should secure all these disperse machines carefully with a firewall.

You may want to read the "deployment" section of the ZeroC manual for more information.

Hope that helps.

Cheers,
~Josh.