Server Overview
What Is a Server?
You can create any number of server instances using MATLAB® Production Server™ software. Each server instance can host any number of deployable archives containing MATLAB code. You may find it helpful to create one server for all archives relating to a particular application. You can also create one server to host code strictly for testing, and so on.
A server instance is considered to be
one unique configuration of the MATLAB Production
Server product. Each configuration has its own options
file (main_config) and diagnostic files (log files,
Process Identification (pid) files, and endpoint files).
In addition, each server has its own auto_deploy folder,
which contains the deployable archives you want the server to host
for clients.
The server also manages the MATLAB Runtime (MATLAB Compiler),
which enables MATLAB code to execute. The settings in main_config determine
how each server interacts with the MATLAB Runtime to process clients
requests. You can set these parameters according to your performance
requirements and other variables in your IT environment.
How Does a Server Manage Work?
A server processes a transaction using these steps:
The client sends MATLAB function calls to the master server process (the main process on the server).
MATLAB function calls are passed to one or more MATLAB Runtime workers.
MATLAB functions are executed by the MATLAB Runtime worker.
Results of MATLAB function execution are passed back to the master server process.
Results of MATLAB function execution are passed back for processing by the client.
The server is the middleman in the MATLAB Production Server™ environment. It simultaneously accepts connections from clients, and then dispatches MATLAB Runtime workers—MATLAB sessions—to process client requests to the MATLAB Runtime. By defining and adjusting the number of workers and threads available to a server, you tune capacity and throughput respectively.
Workers (capacity management) (
num-workers) — The number of MATLAB Runtime workers available to a server.Each worker dispatches one MATLAB execution request to the MATLAB Runtime, interacting with one client at a time. By defining and tuning the number of workers available to a server, you set the number of concurrent MATLAB execution requests that can be processed simultaneously.
num-workersshould roughly correspond to the number of cores available on the local host.Threads (throughput management) (
num-threads) — The number of threads (units of processing) available to the master server process.MATLAB Production Server Data Flow from Client to Server and Back
The server does not allocate a unique thread to each client connection. Rather, when data is available on a connection, the required processing is scheduled on a pool of threads.
--num-threadssets the size of that pool (the number of available request-processing threads) in the master server process. The threads in the pool do not execute MATLAB code directly. Instead, there is a single thread within each MATLAB Runtime worker process that executes MATLAB code on the client's behalf. The number of threads you define to a server should roughly correspond to the number of cores available on the local host.