Back to search

University of Crete (UOC)

Java™ on scalable memory architectures

Abstract

dc:description

The beginning of the new millennium signaled the need for new ways to keep improving the performance of processors. The approach of increasing the frequency as transistors got smaller and smaller is no longer viable due to the increased power-leakage and heat generation. In an effort to maintain the performance increase steady despite the difficulties, processor manufacturers turned to multi-core computing. This way the industry keeps taking advantage of the smaller transistors by packing more compute elements in the same area, rather than making the circuits more complex and increasing the frequency. However, as the number of cores increases new challenges come up. One such challenge regards cache-coherence. Managing hardware caches and keeping the data coherent across them is become more and more complex and energy inefficient as the number of cores grow. To tackle this issue hardware architects are experimenting with modular non cache coherent and partially coherent architectures. Such architectures delegate the memory coherency to the software.In this thesis we study how high productivity languages can be run efficiently on such architectures. High productivity languages, like Java, are designed to abstract away the hardware details and allow developers to focus on the implementation of their algorithm. Such programming languages rely on process virtual machines, like the Java virtual machine, to provide consistent behavior across different platforms. We focus our work on the Java virtual machine since it is one of the most popular and widely studied process virtual machines on top of which tens of languages are being implemented, with the most distinguishable being Java and Scala.Java virtual machine implementations need to adhere to the Java language specifications and the Java memory model. In this thesis we study the Java memory model and present an extension of it that exposes explicit memory transfers between caches. This extension eases the process of arguing about the adherence of a Java virtual machine targeting a non cache coherent architecture by providing explicit rules regarding the ordering of memory transfers in respect to other operations in an Java execution. We sketch the proof that our extension complies to the original Java memory model and allows the same optimizations.We present a Java virtual machine design targeting non cache coherent and partially coherent architectures. Our design aims to minimize the number of memory transfers and messages exchanged while still adhering to the Java memory model. Our design also takes advantage of partial coherence by sharing some structures across different cores on the same coherence island. Based on our design we implement a Java virtual machine and evaluate it on an emulator of a non cache coherent architecture. The results show that our implementation scales up to 500 of cores and its scalability is comparable to that of the state of the art Java virtual machine running on a cache-coherent architecture.Last but not least we model our implementation in the operational semantics of a Java core calculus that we define for this purpose. We then prove that these operational semantics produce only well-formed executions according to the Java memory model. Since the operational semantics model our implementation we argue that the latter also produces only well-formed executions, thus it adheres to the Java memory model.

Degree

thesis:*
Grantor dc:publisher
University of Crete (UOC)
Year dc:date
2016

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Zakkak, Foivos
  • Ζακκάκ, Φοίβος

Subjects

dc:subject × 15

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/40368

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Zakkak, Foivos; Ζακκάκ, Φοίβος. Java™ on scalable memory architectures. University of Crete (UOC), 2016. http://hdl.handle.net/10442/hedi/40368