Eli Tilevich

J-Orchestra: Automatic Java Application Partitioning

J-Orchestra is an automatic partitioning system for Java programs. J-Orchestra takes as input Java applications in bytecode format and transforms them into distributed applications, running on distinct Java Virtual Machines. To accomplish such automatic partitioning, J-Orchestra uses bytecode rewriting to substitute method calls with remote method calls, direct object references with proxy references, etc. Using J-Orchestra does not require great sophistication in distributed system methodology--the user only has to specify the network location of various hardware and software resources and their corresponding application classes. J-Orchestra has significant generality, flexibility, and degree of automation advantages compared to previous work on automatic partitioning. For instance, J-Orchestra can correctly partition almost any pure Java program, allowing any application object to be placed on any machine, regardless of how application objects access each other and Java system objects. This power is due to the novel way that J-Orchestra deals with unmodifiable code (e.g., native code in the Java system classes). Additionally, J-Orchestra offers support for object migration and run-time optimizations, like the lazy creation of distributed objects.We have used J-Orchestra to successfully partition several realistic applications including a command line shell, a ray tracer, and several applications with native dependencies (sound, graphics).

J-Orchestra: Enhancing Java programs with distribution capabilities

J-Orchestra is a system that enhances centralized Java programs with distribution capabilities. Operating at the bytecode level, J-Orchestra transforms a centralized Java program (i.e., running on a single Java Virtual Machine (JVM)) into a distributed one (i.e., running across multiple JVMs). This transformation effectively separates distribution concerns from the core functionality of a program. J-Orchestra follows a semiautomatic transformation process. Through a GUI, the user selects program elements (at class granularity) and assigns them to network locations. Based on the users input, the J-Orchestra backend automatically partitions the program through compiler-level techniques, without changes to the JVM or to the Java Runtime Environment (JRE) classes. By means of bytecode engineering and code generation, J-Orchestra substitutes method calls with remote method calls, direct object references with proxy references, etc. It also translates Java language features (e.g., static methods and fields, inheritance, inner classes, new object construction, etc.) for efficient distributed execution. We detail the main technical issues that J-Orchestra addresses, including its mechanism for program transformation in the presence of unmodifiable code (e.g., in JRE classes) and the translation of concurrency and synchronization constructs to work correctly over the network. We further discuss a case study of transforming a large, commercial, third-party application for efficient execution in a client server environment and outline the architectural characteristics of centralized programs that are amenable to automated distribution with J-Orchestra.

Energy-Efficient and Fault-Tolerant Distributed Mobile Execution

Although battery capacities keep increasing, the execution demands of modern mobile devices continue to outstrip their battery lives. As a result, battery life is bound to remain a key constraining factor in the design of mobile applications. To save battery power, mobile applications are often partitioned to offload parts of their execution to a remote server. However, partitioning an application renders it unusable in the face of network outages. In this paper, we present a novel approach that reduces the energy consumption of mobile applications through server offloading without partitioning. The functionality that consumes energy heavily is executed in the cloud, with the programs state check pointed and transferred across the mobile device and the cloud. Our approach is portable, as it introduces the offloading functionality through byte code enhancement, without any changes to the runtime system. The check pointed states size is minimized through program analysis. In the case of a network outage, the offloading interrupts and the application reverts to executing locally from the latest checkpoint. Our case studies demonstrate how our approach can reduce energy consumption for third-party Android applications. Transformed through our approach, the applications consume between 30% and 60% fewer Joules than their original versions. Our results indicate that portable offloading can improve the battery life of modern mobile applications while maintaining their resilience to network outages.

Annotation refactoring: inferring upgrade transformations for legacy applications

Since annotations were added to the Java language, many frameworks have moved to using annotated Plain Old Java Objects (POJOs) in their newest releases. Legacy applications are thus forced to undergo extensive restructuring in order to migrate from old framework versions to new versions based on annotations (Version Lock-in). Additionally, because annotations are embedded in the application code, changing between framework vendors may also entail largescale manual changes (Vendor Lock-in). This paper presents a novel refactoring approach that effectively solves these two problems. Our approach infers a concise set of semantics-preserving transformation rules from two versions of a single class. Unlike prior approaches that detect only simple structural refactorings, our algorithm can infer general composite refactorings and is more than 97% accurate on average. We demonstrate the effectiveness of our approach by automatically upgrading more than 80K lines of the unit testing code of four open-source Java applications to use the latest version of the popular JUnit testing framework.

Reducing the Energy Consumption of Mobile Applications Behind the Scenes

As energy efficiency has become a key consideration in the engineering of mobile applications, an increasing number of perfective maintenance tasks are concerned with optimizing energy consumption. However, optimizing a mobile application to reduce its energy consumption is non-trivial due to the highly volatile nature of mobile execution environments. Mobile applications commonly run on a variety of mobile devices over mobile networks with divergent characteristics. Therefore, no single, static energy consumption optimization is likely to yield across-the-board benefits, and may even turn to be detrimental in some scenarios. In this paper, we present a novel approach to perfective maintenance of mobile applications to reduce their energy consumption. The maintenance programmer declaratively specifies the suspected energy consumption hotspots in a mobile application. Based on this input, our approach then automatically transforms the application to enable it to offload parts of its functionality to the cloud. The offloading is highly adaptive, being driven by a runtime system that dynamically determines both the state-to-offload and its transfer mechanism based on the execution environment in place. In addition, the runtime system continuously improves its effectiveness due to a feedback-loop mechanism. Thus, our approach flexibly reduces the energy consumption of mobile applications behind the scenes. Applying our approach to third- party Android applications has shown that it can effectively reduce the overall amount of energy consumed by these applications, with the actual numbers ranging between 25% and 50%. These results indicate that our approach represents a promising direction in developing pragmatic and systematic tools for the perfective maintenance of mobile applications.

To upgrade or not to upgrade: impact of online upgrades across multiple administrative domains

Online software upgrades are often plagued by runtime behaviors that are poorly understood and difficult to ascertain. For example, the interactions among multiple versions of the software expose the system to race conditions that can introduce latent errors or data corruption. Moreover, industry trends suggest that online upgrades are currently needed in large-scale enterprise systems, which often span multiple administrative domains (e.g., Web 2.0 applications that rely on AJAX client-side code or systems that lease cloud-computing resources). In such systems, the enterprise does not control all the tiers of the system and cannot coordinate the upgrade process, making existing techniques inadequate to prevent mixed-version races. In this paper, we present an analytical framework for impact assessment, which allows system administrators to directly compare the risk of following an online-upgrade plan with the risk of delaying or canceling the upgrade. We also describe an executable model that implements our formal impact assessment and enables a systematic approach for deciding whether an online upgrade is appropriate. Our model provides a method of last resort for avoiding undesirable program behaviors, in situations where mixed-version races cannot be avoided through other technical means.

NRMI: natural and efficient middleware

We present NRMI: a drop-in replacement of Java RMI that offers call-by-copy-restore semantics for arbitrary linked data structures, in addition to regular call-by-copy semantics. Call-by-copy-restore middleware is more natural to use than traditional call-by-copy RPC mechanisms, enabling distributed calls to behave much like local calls. We discuss in depth the effect of calling semantics for middleware, describe how call-by-copy-restore can be implemented efficiently, and show examples of Java programs where NRMI is more convenient than regular Java RMI.We present natural remote method invocation (NRMI): a middleware mechanism that provides a fully general implementation of call-by-copy-restore semantics for arbitrary linked data structures, used as parameters in remote procedure calls. Call-by-copy-restore offers a more natural programming model for distributed systems than traditional call-by-copy middleware, enabling remote calls to behave much like local calls. We discuss in depth the effects of calling semantics for middleware, describe when and why NRMI is more convenient to use than standard middleware, and present three implementations of NRMI in distinct settings, showing the generality of the approach.

Cloud refactoring: automated transitioning to cloud-based services

Using cloud-based services can improve the performance, reliability, and scalability of a software application. However, transitioning an application to use cloud-based services is difficult, costly, and error-prone. The required re-engineering effort includes migrating to the cloud the functionality to be accessed as remote cloud-based services and re-targeting the client code accordingly. In addition, the client must be able to detect and handle the faults raised in the process of invoking the services. As a means of streamlining this transitioning, we developed a set of refactoring techniques—automated, IDE-assisted program transformations that eliminate the need to change programs by hand. In particular, we show how a programmer can extract services, add fault tolerance functionality, and adapt client code to invoke cloud services via refactorings integrated with a modern IDE. As a validation, we have applied our approach to automatically transform two third-party Java applications to use cloud-based services. We have also applied our approach to re-engineer a suite of services operated by General Electric to use cloud-based resources to better satisfy the GE business requirements.

Remote Batch Invocation for Compositional Object Services

Because Remote Procedure Calls do not compose efficiently, designers of distributed object systems use Data Transfer and Remote Facade patterns to create large-granularity interfaces, hard-coded for particular client use cases. As an alternative to RPC-based distributed objects, this paper presents Remote Batch Invocation (RBI), language support for explicit client-defined batches. A Remote Batch statement combines remote and local execution: all the remote code is executed in a single round-trip to the server, where all data sent to the server and results from the batch are communicated in bulk. RBI supports remote blocks, iteration and conditionals, and local handling of remote exceptions. RBI is efficient even for fine-grained interfaces, eliminating the need for hand-optimized server interfaces. We demonstrate RBI with an extension to Java, using RMI internally as the transport layer. RBI supports large-granularity, stateless server interactions, characteristic of service-oriented computing.

Portable and efficient distributed threads for Java

Java middleware mechanisms, such as Java RMI or CORBA implementations, do not support thread coordination over the network: synchronizing on remote objects does not work correctly and thread identity is not preserved for executions spanning multiple machines. The current approaches dealing with the problem suffer from one of two weaknesses: either they require a new middleware mechanism, making them less portable, or they add overhead to the execution to propagate a thread identifier through all method calls. In this paper we present an approach that works with an unmodified middleware implementation, yet does not impose execution overhead. The key to our technique is the bytecode transformation of only stub routines, instead of the entire client application. We argue that this approach is portable and can be applied to mostly any middleware mechanism. At the same time, we show that, compared to past techniques, our approach eliminates an overhead of 5.5-12 applications from the SPEC JVM suite.