Location
Grand Wailea, Hawaii
Event Website
https://hicss.hawaii.edu/
Start Date
7-1-2020 12:00 AM
End Date
10-1-2020 12:00 AM
Description
An important issue for concurrent garbage collection in virtual machines (VM) is to identify which garbage collector (GC) to use during the collection process. For instance, Java program execution times differ greatly based on the employed GC. It has not been possible to identify the optimal GC algorithms for a specific program before exhaustively profiling the execution times for all available GC algorithms. In this paper, we present an adaptive and concurrent garbage collection (ACGC) technique that can predict the optimal GC algorithm for a program without going through all the GC algorithms. We implement this technique in the Java virtual machine and test it using standard benchmark suites. ACGC learns the algorithms’ usage pattern from different training program features and generates a model for future programs. Feature generation and selection are two important steps of our technique, which creates different attributes to use in the learning step. Our experimental evaluation shows improvement in selecting the best GC. Additionally, our approach is helpful in finding better heap size settings for improved program execution.
Adaptive and Concurrent Garbage Collection for Virtual Machines
Grand Wailea, Hawaii
An important issue for concurrent garbage collection in virtual machines (VM) is to identify which garbage collector (GC) to use during the collection process. For instance, Java program execution times differ greatly based on the employed GC. It has not been possible to identify the optimal GC algorithms for a specific program before exhaustively profiling the execution times for all available GC algorithms. In this paper, we present an adaptive and concurrent garbage collection (ACGC) technique that can predict the optimal GC algorithm for a program without going through all the GC algorithms. We implement this technique in the Java virtual machine and test it using standard benchmark suites. ACGC learns the algorithms’ usage pattern from different training program features and generates a model for future programs. Feature generation and selection are two important steps of our technique, which creates different attributes to use in the learning step. Our experimental evaluation shows improvement in selecting the best GC. Additionally, our approach is helpful in finding better heap size settings for improved program execution.
https://aisel.aisnet.org/hicss-53/da/soft_computing/3