Technical Sessions

This paper tackles the problem of providing familiar OS abstractions for I/O (such as pipes, network sockets, and a shared file system) to applications on heterogeneous cores including accelerators, co-processors, and ooad engines. We aim to isolate the implementation of these facilities from the details of a platform’s memory architecture, which is likely to include a combination of cache-coherent shared memory, non-cache-coherent shared memory, and non-shared memory, all in the same system.

We propose coherence-oblivious sharing (Cosh), a new OS abstraction that provides inter-process sharing with clear semantics on such diverse hardware. We have implemented a prototype of Cosh for the Barrelfish multikernel. We describe how to build common OS functionality using Cosh, and evaluate its performance on a heterogeneous system consisting of commodity cache-coherent CPUs and prototype Intel many-core co-processors.

We introduce Fracture, a novel framework that transforms and modernizes the basic process abstraction. By “fracturing” an application into logical modules, Fracture enables powerful and novel run-time configurations that improve run-time testing, application availability, and general robustness, all in a generic and incremental manner. We demonstrate the utility of fracturing via in-depth case studies of a chat client, a web server, and two user-level file systems. Through these examples, we show that Fracture enables applications to transparently tolerate memory leaks, buffer overflows, and isolate subsystem crashes, with little change to source code; through intelligent fracturing, we can achieve low overhead as well, thus enabling deployment.

Optimization of application and system software for energy efficiency is of ecological, economical, and technical importance—and still challenging. Deficiency in adequate tooling support is a major issue. The few tools available (i.e., measurement instruments, energy profilers) have poorly conceived interfaces and their integration into widely used development processes is missing. This implies time-consuming, tedious measurements and profiling runs and aggravates, if not shoots down, the development of energy-efficient software.

We present PEEK, a systems approach to proactive energy-aware programming. PEEK fully automates energy measurement tasks and suggests program-code improvements at development time by providing automatically generated energy optimization hints. Our approach is based on a combined software and hardware infrastructure to automatically determine energy demand of program code and pinpoint energy faults, thereby integrating seamlessly into existing software development environments. As part of PEEK we have designed a lightweight, yet powerful electronic measuring device capable of taking automated, analog energy measurements. Results show an up to 8.4-fold speed-up of energy analysis when using PEEK, while the energy consumption of the analyzed code was improved by 25.3%.

Mobile devices regularly move between feast and famine—environments that differ greatly in the capacity and cost of available network resources. Managing these resources effectively is an important aspect of a user’s mobile experience. However, preferences for resource management vary across users, time, and operating conditions, and user and application interests may not align. Furthermore, today’s mobile OS mechanisms are typically coarse-grained, inflexible, and scattered across system and application settings. Users must adopt a "one size fits all" solution or micro-manage their devices.

This paper introduces Tango, a platform for managing network resource usage through a programmatic model that expresses user and app interests (“policies”). Tango centralizes policy expression and enforcement in a controller process that monitors device state and adjusts network usage according to a user’s (potentially dynamic) interests. To align interests and leverage app-specific knowledge, Tango uses a constraint model that informs apps of network limitations so they can optimize their usage. We evaluate how to design policies that account for data limits, user experience, and battery life. We demonstrate how Tango improves individual network-intensive apps like music streaming, as well as conditions when multiple apps compete for limited resources.

In this paper, we describe an experimental study of very long response time (VLRT) requests in the latency long tail problem. Applying micro-level event analysis on fine-grained measurement data from n-tier application benchmarks, we show that very short bottlenecks (from tens to hundreds of milliseconds) can cause queue overflows that propagate through an n-tier system, resulting in dropped messages and VLRT requests due to timeout and retransmissions. Our study shows that even at moderate CPU utilization levels, very short bottlenecks arise from several system layers, including Java garbage collection, anti-synchrony between workload bursts and DVFS clock rate adjustments, and statistical workload interferences among co-located VMs.

As a simple model for a variety of causes of VLRT requests, very short bottlenecks form the basis for a discussion of general remedies for VLRT requests, regardless of their origin. For example, methods that reduce or avoid queue amplification in an n-tier system result in non-trivial trade-offs among system components and their configurations. Our results show interesting challenges remain in both causes and effective remedies of very short bottlenecks.

In the age of cloud computing, securely storing, tracking, and controlling access to digital “secrets” (e.g. private cryptographic keys, hashed passwords, etc) is a major challenge for developers, administrators, and end-users alike. Yet, the ability to securely store such secrets is critical to the security of the web-connected applications on which we rely. We believe many of the traditional challenges to the secure storage of digital secrets can be overcome through the creation of a dedicated “Secret Storage as a Service” (SSaaS) interface. Such an interface allows us to separate secure secret storage and access control from the applications that require such services. We present Custos: an SSaaS prototype. We describe the Custos design principles and architecture. We also discuss a range of applications in which Custos can be leveraged to store secrets such as cryptographic keys. We compare Custos-backed versions of such applications to the existing alternatives and discuss how Custos and the SSaaS model can improve the security of such applications while still supporting the wide range of features (e.g. multi-device syncing, multi-user sharing, etc) we have come to expect in the age of the Cloud.