MATLAB®
Production Server™ instances can host deployable archives
compiled using multiple versions of MATLAB
Compiler SDK™. You configure
a server instance to do this by adding multiple mcr-root properties
to the configuration file for the instance:
Install the required versions of the MATLAB Runtime.
A server instance should only be configured to use MATLAB Runtime roots on a local file system. Otherwise, a network partition may cause worker processes to fail.
All values for mcr-root must be
for the same OS/hardware combination.
If the server instance is running, stop it.
Open the configuration file for the instance in a text editor.
The configuration file is at instanceRoot/config/main_config
Locate the entry for the mcr-root property.
--mcr-root mCRuNsETtOKEN
For each version of the MATLAB Runtime the instance
supports, add an instance of the mcr_root property.
For example, to configure the instance to use the v81 and v82 versions of the MATLAB Runtime.
--mcr-root C:\Program Files\MATLAB\MATLAB Compiler Runtime\v81 --mcr-root C:\Program Files\MATLAB\MATLAB Runtime\v82
Restart the server instance.
Once the server instance is configured to use multiple versions of MATLAB Runtime, it scans the list of provided MATLAB Runtime installations in order from first to last and chooses the first MATLAB Runtime installation capable of processing the request. A MATLAB Runtime installation can process a request if it is compatible with the version of MATLAB used to create the deployable archive containing the function being evaluated.
Since the server instance always chooses the first compatible version of MATLAB Runtime, configuring the server instance with multiple instances of the same MATLAB Runtime version has no effect on performance.
Configuring a server instance to use multiple MATLAB Runtime versions also changes how the server instance manages the workers used to process requests.
When using a single MATLAB Runtime installation, the server
instance starts workers as needed until num-workers workers
are running. Once running, workers maybe restarted in response to
the worker-restart-interval property
or the worker-restart-memory-limit property.
Workers are never fully stopped.
Once a server instance starts using multiple MATLAB Runtime versions,
it dynamically manages the worker pool. The server instance starts
new workers as needed until num-workers workers
are running. The worker instances are spread out over the different MATLAB Runtime versions.
Once num-workers workers
are running, the server instance returns workers to the pool of available
workers based on the worker-memory-trigger property
and the queue-time-trigger property. Once worker
is returned to the pool, it can be allocated to process new requests
using any of the configured MATLAB Runtime versions.