The avant-garde of high performance computing is building petabyte storage systems. At CITI, we are investigating the use of NFSv4 as a standard for fast and secure access to this data, both across a WAN and within a (potentially massive) cluster. An NFSv4 server manages much state information, which hampers exporting objects via multiple servers and allows the NFSv4 server to become a bottleneck as load increases. This paper introduces Parallel NFSv4, extending the NFSv4 protocol with a new server-to-server protocol and a new

Today’s batch computing systems derive from a job scheduling paradigm. In contrast, utility service environments assign measured quantities of resources to long-running customer activities according to policies for resource matching and allocation. In this paper we explore the duality between job scheduling and utility resource management using a leasing abstraction that generalizes reservations. We suggest that advanced job management systems will increasingly incorporate elements of utility systems to support new capabilities for urgent computing, predictable execution of large composite workflows, manageability and customization, and technology trends such as server virtualization. We show that similar resource sharing policies may be implemented using either abstraction, and explore a design space of hybrid systems that combine leases and job scheduling to balance various objectives. Experiments compare and characterize representative models from the design space in terms of job scheduling and resource management. 1

VFS lookup code examines and translates path names one component at a time, checking for special cases such as mount points and symlinks. VFS calls the NFS lookup operation as necessary. NFS employs caching to reduce the number of lookup operations that go to the server. However, when part or all of a path is not cached, NFS lookup operations go back to the server. Although NFS's caching is effective, component-bycomponent translation of an uncached path is inefficient, enough so that lookup is typically the operation most commonly processed by servers. We study the effect of augmenting the VFS lookup algorithm and the NFS protocol so that a client can ask a server to translate an entire path in a single operation. The preconditions for a successful request are usually but not always satisfied, so the algorithm is optimistic. This small change can deliver substantial improvements in client latency and server load. 1 Introduction The NFS lookup operation frequently goes "over the wire...

We describe two modifications to the FreeBSD 4.6 NFS server to increase read throughput by improving the read-ahead heuristic to deal with reordered requests and stride access patterns. We show that for some stride access patterns, our new heuristics improve end-to-end NFS throughput by nearly a factor of two. We also show that benchmarking and experimenting with changes to an NFS server can be a subtle and challenging task, and that it is often difficult to distinguish the impact of a new algorithm or heuristic from the quirks of the underlying software and hardware with which they interact. We discuss these quirks and their potential effects.

A Systematic Characterization of Application Sensitivity to Network Performance by Richard Paul Martin Doctor of Philosophy in Computer Science University of California at Berkeley Professor David E. Culler, Chair This thesis provides a systematic study of application sensitivity to network performance. Our aim is to investigate the impact of communication performance on real applications. Using the LogGP model as an abstract framework, we set out to understand which aspects of communication performance are most important. The focus of our investigation thus centers on a quantification of the sensitivityof applications to the parameters of the LogGP model: network latency, software overhead, per-message and per-byte bandwidth. We define sensitivity as the change in some application performance metric, such as run time or updates per second, as a function of the LogGP parameters. The strong association of the LogGP model with real machine components allows us to draw architectural conclusions from the measured sensitivity curves as well.

while a modification that fills a block results in an immediate, although asynchronous, write. The Some file systems can delay writing modified ULTRIX operating system can be configured to be data to disk, in order to reduce disk traffic and overeven more aggressive, delaying all writes of modified head. Prudence dictates that such delays be bounded, data. in case the system crashes. We refer to an algorithm used to decide when to write delayed data back to Without some bound on the age of a delayeddisk as an update policy. Traditional UNIX systems write block, a system crash could cause loss of aruse a periodic update policy, writing back all bitrary data. Users would not tolerate this, so the file delayed-write data once every 30 seconds. Periodic system does push delayed-write data out to disk, after update is easy to implement but performs quite badly a while. We use the term update policy to describe in some cases. This paper describes an approximate the algorithm that decides what to write out, and implementation of an interval periodic update policy, when.