Study of Compilation and Heap Tuning Parameters across Virtual Machines for the Application Performance Improvements

Virtual Machines are the programs for supporting platform independence, network mobility and scalability for different underlying machine architectures such as x86 and ARM etc for JAVA based framework. VM main functions are Compilation of bytecode to the native code, Memory Management and Garbage Collector Management etc. Java architecture is a stack based architecture where as Dalvik and ART possesses register based architecture. All mentioned Virtual Machines are run on battery driven devices such as laptop and mobile. So the goal of VM must be to improve performance by consuming less battery power and faster execution of programs. It is also important to maximize the throughput of the CPU cycles. CPU cycles are one more major factor to consider which impact the battery life. Many enhancements have been observed in VM evolution from JAVA VM to ART VM.JAVA Code compilation has been taken care by interpreter and just in time compilation technique which convert per java class to native code where as dalvik bytecode per application forms Dex executable and further conversion of Dex to native by using JIT compilation in runtime. ART uses new Ahead of Time compilation technique which creates executable and linkable format. ART compilation done during boot or upgrade time and overcome the runtime compilation. So far ART Virtual Machine has been considered as best VM in terms of performance and better user experience for android system under command of linux operating system. Tradeoffs took place for performance over higher installation time for application in case of ART (Android Runtime) [14, 15, 16, 17]. Virtual Machines also evolved and experimented with different memory management techniques for fast allocation and aim to achieve parallelism and concurrency. Allocation technique used for JAVA and Dalvik VM is through dlmalloc library whereas rosalloc technique used for Android Runtime to boost performance in allocation [2, 5].Garbage collection runs as daemon in each VM for reclaiming the memory and causes application performance to go low. Several GC techniques have been experimented on VM’s among which Concurrent Mark and Sweep and Generational are most famous one. Goal of GC must be lesser pause time, GC time and to overcome UI frame drops [1, 13]. This paper presents different compilation and heap tuning parameters for different virtual machine. Experimentation results discuss about the trade-off with tuning.