HFSS15: Integrating ANSYS EM Tools with Third Party Schedulers

This document indicates how to create a dynamically linked library to allow integration of ANSYS EM tools with an arbitrary scheduler environment. Each scheduler proxy library is used for a single specific scheduler environment. If the library is installed with a valid name and in the correct location, then it will automatically be loaded and used by ANSYS EM tools.

• Introduction

• Common Requirements for Running Jobs

• Using a Shared Library (Linux) or a DLL (Microsoft Windows)

• Scheduler Proxy Interfaces

• Using an IronPython Program for Scheduler Integration

Introduction

ANSYS EM Software Tools may be run as serial or parallel jobs on a cluster under control of a scheduler. Serial jobs are run using a single analysis engine at any one time on a single host. If the tool performs multiple analyses (for a frequency sweep or a parametric analysis, for example), the analyses are performed one after the other. Parallel jobs are run using multiple analysis engines running in parallel on the same host or on separate hosts. For parts of the analysis (such as meshing), the parallel job may use only a single analysis engine on a single host. Other parts of the analysis (such as a frequency sweep, parametric analysis or DDM, for example) may be distributed to multiple analysis engines running in parallel.

• Serial Jobs

• Parallel Jobs

Serial Jobs

When an ANSYS EM batch analysis runs as a serial job, the analysis engines run on the same host as the desktop process. The desktop process does not need to interact with the scheduler to get the names of hosts allocated to the job or to start processes on other hosts.

Parallel Jobs

For a parallel job, the desktop process starts multiple analysis engines that run in parallel. These engines may be started on the host where the desktop process is running, or on other hosts allocated to the job. The desktop process interacts with the scheduler to obtain information on the hosts that are allocated to the job, and to start engines on the local host or on other hosts allocated to the job. This document provides information on how to facilitate this interaction between the desktop process and the scheduler controlling the cluster.

For some popular job schedulers in a standard configuration, ANSYS EM provides an "out of the box" integrated solution that will work with the scheduler. In this case, the ANSYS EM installation includes code that will determine if the analysis is running as a scheduler job and communicate with the scheduler when needed. For other schedulers, the code to obtain information about the hosts allocated to a job and to distribute portions of the job to hosts assigned to the job is not provided in the installation. In order to facilitate using ANSYS EM Software Tools with other schedulers, the user may provide a way for ANSYS EM Tools to interact with the scheduler. Currently, two general approaches are available to users.

In the first approach, the user creates a shared library (on Linux) or a dynamically linked library (on Microsoft Windows) to provide communication between the ANSYS EM Tool and the scheduler. This library is loaded by the ANSYS EM Tool at runtime, and if the ANSYS EM Tool is running as part of a scheduler job, the ANSYS EM Tool interacts with the library to get information from the scheduler, and to start additional processes on specified hosts. Each such library implements the same set of extern "C" functions needed to mediate the interactions between the ANSYS EM Tool and the scheduler. The details of these functions are described below.

In the second approach, the user creates an IronPython program to provide communication between the ANSYS EM Tool and the scheduler. This program is loaded by the ANSYS EM Tool at runtime, and if the ANSYS EM Tool is running as part of a scheduler job, the ANSYS EMs Tool uses the IronPython program to get information from the scheduler, and to to start additional processes on specified hosts. Each python script contains a class implementing a specified interface, which contains functions needed to mediate the interactions between the ANSYS EM Tool and the scheduler. The details of the interface are described below. The IronPython interface is equivalent to the extern "C" functions used in the first approach.