Margo I. Seltzer

The case for geographical push-caching

Most wide-area caching schemes are client initiated. Decisions on when and where to cache information are made without the benefit of the servers global knowledge of the situation. We believe that the server should play a role in making these caching decisions, and we propose geographical push-caching as a way of bringing the server back into the loop. The World Wide Web is an excellent example of a wide-area system that will benefit from geographical push-caching, and we present an architecture that allows a Web server to autonomously replicate HTML pages.

Network-Aware Operator Placement for Stream-Processing Systems

To use their pool of resources efficiently, distributed stream-processing systems push query operators to nodes within the network. Currently, these operators, ranging from simple filters to custom business logic, are placed manually at intermediate nodes along the transmission path to meet application-specific performance goals. Determining placement locations is challenging because network and node conditions change over time and because streams may interact with each other, opening venues for reuse and repositioning of operators. This paper describes a stream-based overlay network (SBON), a layer between a stream-processing system and the physical network that manages operator placement for stream-processing systems. Our design is based on a cost space, an abstract representation of the network and on-going streams, which permits decentralized, large-scale multi-query optimization decisions. We present an evaluation of the SBON approach through simulation, experiments on PlanetLab, and an integration with Borealis, an existing stream-processing engine. Our results show that an SBON consistently improves network utilization, provides low stream latency, and enables dynamic optimization at low engineering cost.

Dealing with disaster: surviving misbehaved kernel extensions

Today’s extensible operating systems allow applications to modify kernel behavior by providing mechanisms for application code to run in the kernel address space. The advantage of this approach is that it provides improved application flexibility and performance; the disadvantage is that buggy or malicious code can jeopardize the integrity of the kernel. It has been demonstrated that it is feasible to use safe languages, software fault isolation, or virtual memory protection to safeguard the main kernel. However, such protection mechanisms do not address the full range of problems, such as resource hoarding, that can arise when application code is introduced into the kernel. In this paper, we present an analysis of extension mechanisms in the VINO kernel. VINO uses software fault isolation as its safety mechanism and a lightweight transaction system to cope with resource-hoarding. We explain how these two mechanisms are sufficient to protect against a large class of errant or malicious extensions, and we quantify the overhead that this protection introduces. We find that while the overhead of these techniques is high relative to the cost of the extensions themselves, it is low relative to the benefits that extensibility brings.

Non-volatile memory for fast, reliable file systems

Given the decreasing cost of non-volatile RAM (NVRAM), by the late 1990’s it will be feasible for most workstations to include a megabyte or more of NVRAM, enabling the design of higher-performance, more reliable systems. We present the trace-driven simulation and analysis of two uses of NVRAM to improve I/O performance in distributed file systems: non-volatile file caches on client workstations to reduce write traffic to file servers, and write buffers for write-optimized file systems to reduce server disk accesses. Our results show that a megabyte of NVRAM on diskless clients reduces the amount of file data written to the server by 40 to 50%. Increasing the amount of NVRAM shows rapidly diminishing returns, and the particular NVRAM block replacement policy makes little difference to write traffic. Closely integrating the NVRAM with the volatile cache provides the best total traffic reduction. At today’s prices, volatile memory provides a better performance improvement per dollar than NVRAM for client caching, but as volatile cache sizes increase and NVRAM becomes cheaper, NVRAM will become cost effective. On the server side, providing a one-half megabyte write-buffer per file system reduces disk accesses by about 20% on most of the measured logstructured file systems (LFS), and by 90% on one heavilyused file system that includes transaction-processing workloads.

Improving Performance Isolation on Chip Multiprocessors via an Operating System Scheduler

We describe a new operating system scheduling algorithm that improves performance isolation on chip multiprocessors (CMP). Poor performance isolation occurs when an applications performance is determined by the behaviour of its co-runners, i.e., other applications simultaneously running with it. This performance dependency is caused by unfair, co- runner-dependent cache allocation on CMPs. Poor performance isolation interferes with the operating system s control over priority enforcement and hinders QoS provisioning. Previous solutions required modifications to the hardware. We present a new software solution. Our cache-fair algorithm ensures that the application runs as quickly as it would under fair cache allocation, regardless of how the cache is actually allocated. If the thread executes fewer instructions per cycle than it would under fair cache allocation, the scheduler increases that threads CPU time slice. This way, the threads overall performance does not suffer because it is allowed to use the CPU longer. We describe our implementation of the algorithm in Solaristrade 10, and show that it significantly improves performance isolation for SPEC CPU, SPEC JBB and TPC-C.

Operating system benchmarking in the wake of lmbench : a case study of the performance of NetBSD on the Intel x86 architecture

The lmbench suite of operating system microbenchmarks provides a set of portable programs for use in cross-platform comparisons. We have augmented the lmbench suite to increase its flexibility and precision, and to improve its methodological and statistical operation. This enables the detailed study of interactions between the operating system and the hardware architecture. We describe modifications to lmbench, and then use our new benchmark suite, hbench:OS, to examine how the performance of operating system primitives under NetBSD has scaled with the processor evolution of the Intel x86 architecture. Our analysis shows that off-chip memory system design continues to influence operating system performance in a significant way and that key design decisions (such as suboptimal choices of DRAM and cache technology, and memory-bus and cache coherency protocols) can essentially nullify the performance benefits of the aggressive execution core and sophisticated on-chip memory system of a modern processor such as the Intel Pentium Pro.

File system aging—increasing the relevance of file system benchmarks

Benchmarks are important because they provide a means for users and researchers to characterize how their workloads will perform on different systems and different system architectures. The field of file system design is no different from other areas of research in this regard, and a variety of file system benchmarks are in use, representing a wide range of the different user workloads that may be run on a file system. A realistic benchmark, however, is only one of the tools that is required in order to understand how a file system design will perform in the real world. The benchmark must also be executed on a realistic file system. While the simplest approach may be to measure the performance of an empty file system, this represents a state that is seldom encountered by real users. In order to study file systems in more representative conditions, we present a methodology for aging a test file system by replaying a workload similar to that experienced by a real file system over a period of many months, or even years. Our aging tools allow the same aging workload to be applied to multiple versions of the same file system, allowing scientific evaluation of the relative merits of competing file system designs.In addition to describing our aging tools, we demonstrate their use by applying them to evaluate two enhancements to the file layout policies of the UNIX fast file system.

The case for application-specific benchmarking

Most performance analysis today uses either microbenchmarks or standard macrobenchmarks (e.g. SPEC, LADDIS, the Andrew benchmark). However, the results of such benchmarks provide little information to indicate how well a particular system will handle a particular application. Such results are, at best, useless and, at worst, misleading. In this paper we argue for an application-directed approach to benchmarking, using performance metrics that reflect the expected behavior of a particular application across a range of hardware or software platforms. We present three different approaches to application-specific measurement, one using vectors that characterize both the underlying system and an application, one using trace-driven techniques, and a hybrid approach. We argue that such techniques should become the new standard.

Provenance: a future history

Science, industry, and society are being revolutionized by radical new capabilities for information sharing, distributed computation, and collaboration offered by the World Wide Web. This revolution promises dramatic benefits but also poses serious risks due to the fluid nature of digital information. One important cross-cutting issue is managing and recording provenance, or metadata about the origin, context, or history of data. We posit that provenance will play a central role in emerging advanced digital infrastructures. In this paper, we outline the current state of provenance research and practice, identify hard open research problems involving provenance semantics, formal modeling, and security, and articulate a vision for the future of provenance.

Self-monitoring and self-adapting operating systems

Extensible operating systems allow applications to modify kernel behavior by providing mechanisms for application code to run in the kernel address space. Extensibility enables a system to efficiently support a broader class of applications than is currently supported. This paper discusses the key challenge in making extensible systems practical: determining which parts of the system need to be extended and how. The determination of which parts of the system need to be extended requires self-monitoring, capturing a significant quantity of data about the performance of the system. Determining how to extend the system requires self-adaptation. In this paper, we describe how an extensible operating system (VINO) can use in situ simulation to explore the efficacy of policy changes. This automatic exploration is applicable to other extensible operating systems and can make these systems self-adapting to workload demands.