METHOD FOR DYNAMIC COMPUTING RESOURCE MANAGEMENT IN MR SYSTEMS
Publication Date: 2017-May-09
The IP.com Prior Art Database
A method for designing MR systems with shared resources such that the total quanta of resources is less than the quanta of resources deployed cumulatively in different subsystems of conventional MR systems, and allocating resources dynamically based on performance priority to achieve desired performance.
The present disclosure relates generally to magnetic resonance (MR) imaging systems, and more particularly to computing resource management in MR systems.
In MR systems, most resource (CPU processing power, memory) intensive computing resources include the Host module for UI/workflow, which needs to support a large number of serials in a single exam, Recon module for reconstruction, and Post Processing module for DICOM post processing.
To achieve desired performance, separate computing machines (servers or workstations) are used for Recon modules, which makes the system architecture complex, and drives up the costs. Recon performance is particularly sensitive to the availability of CPU processing power and Memory. Host and Viz performance is similarly more sensitive with respect to Memory, while the sensitivity to the CPU processing power is lower compared to Recon. However, in current MR systems, most of the pre-defined computing resources are idle at most of time (such as during Recon).
Therefore, there exists a need to decrease the computing platform cost while providing acceptable performance.
BRIEF DESCRIPTION OF DRAWINGS
Figure 1 depicts a schematic illustrating a comparison between CPU management strategy under contrasting loads of a normal scan and a stress scan with a high Recon load.
The present disclosure employs a flexible approach to share computing resource in each of the MR subsystems, and particularly, achieve a target configuration having a smaller resource footprint than the combined footprint of conventional Host and Recon modules.
For example, the techniques comprise sharing the CPU, Memory, and Storage resources across the Host, Recon, and Viz modules, and utilizing a single workstation (or server) for all subsystems. The MR system does not allocate extensive resources to the Recon module when a scan is not active. Such dynamic resource allocation of resources away from the Recon module is based on channel number and application type. For example, in a Linux implementation, use ‘cgroup’ to implement the dynamic resource allocation, including CPU, Memory, network, IO(disk).
As illustrated in FIG. 1, different strategies...