Using Speculation to Reduce Server Load and Service Time on the WWW

Speculative service implies that a client's request for a document is serviced by sending, in addition to the document requested, a number of other documents (or pointers thereto) that the server speculates will be requested by the client in the near future. This speculation is based on statistical information that the server maintains for each document it serves. The notion of speculative service is analogous to prefetching, which is used to improve cache performance in distributed/parallel shared memory systems, with the exception that servers (not clients) control when and what to prefetch. Using trace simulations based on the logs of our departmental HTTP server http://cs-www.bu.edu, we show that both server load and service time could be reduced considerably, if speculative service is used. This is above and beyond what is currently achievable using client-side caching [3] and server-side dissemination [2]. We identify a number of parameters that could be used to fine-tune the level of speculation performed by the server.

Dynamic Server Selection in the Internet

As distributed information services like the World Wide Web become increasingly popular on the Internet, problems of scale are clearly evident. A promising technique that addresses many of these problems is service (or document) replication. However, when a service is replicated, clients then need the additional ability to find a "good" provider of that service. In this paper we report on techniques for finding good service providers without a priori knowledge of server location or network topology. We consider the use of two principal metrics for measuring distance in the Internet: hops, and round-trip latency. We show that these two metrics yield very different results in practice. Surprisingly, we show data indicating that the number of hops between two hosts in the Internet is not strongly correlated to round-trip latency. Thus, the distance in hops between two hosts is not necessarily a good predictor of the expected latency of a document transfer. Instead of using known or measured distances in hops, we show that the extra cost at runtime incurred by dynamic latency measurement is well justified based on the resulting improved performance. In addition we show that selection based on dynamic latency measurement performs much better in practice that any static selection scheme. Finally, the difference between the distribution of hops and latencies is fundamental enough to suggest differences in algorithms for server replication. We show that conclusions drawn about service replication based on the distribution of hops need to be revised when the distribution of latencies is considered instead.

Dynamic Server Selection using Bandwidth Probing in Wide-Area Networks

Replication is a commonly proposed solution to problems of scale associated with distributed services. However, when a service is replicated, each client must be assigned a server. Prior work has generally assumed that assignment to be static. In contrast, we propose dynamic server selection, and show that it enables application-level congestion avoidance. To make dynamic server selection practical, we demonstrate the use of three tools. In addition to direct measurements of round-trip latency, we introduce and validate two new tools: bprobe, which estimates the maximum possible bandwidth along a given path; and cprobe, which estimates the current congestion along a path. Using these tools we demonstrate dynamic server selection and compare it to previous static approaches. We show that dynamic server selection consistently outperforms static policies by as much as 50%. Furthermore, we demonstrate the importance of each of our tools in performing dynamic server selection.

Performance Analysis of a WWW Server

This paper addresses the problem of analyzing performance of WWW servers. The web has experienced a phenomenal growth and has become the most popular Internet application. As a consequence of its large popularity, the Internet has suffered from various performance problems, such as network congestion and overloaded servers. These days, it is not uncommon to find servers refusing connections because they are overloaded. Performance has always been a key issue in the design and operation of on-line systems. With regard to Internet, performance is also critical, because users want fast and easy access to all objects (i.e., documents, pictures, audio, and video) available on the net. Thus, it is important to understand WWW performance issues. This paper focuses on the performance analysis of a Web server. Using a synthetic benchmark (WebStone), we analyze three different Web server software running on top of a Windows NT platform and performing some typical WWW tasks.

Measuring the Behavior of a World-Wide Web Server

Server performance has become a crucial issue for improving the overall performance of the World-Wide Web. This paper describes Webmonitor, a tool for evaluating and understanding server performance, and presents new results for a realistic workload. Webmonitor measures activity and resource consumption, both within the kernel and in HTTP processes running in user space. Webmonitor is implemented using an efficient combination of sampling and event-driven techniques that exhibit low overhead. Our initial implementation is for the Apache World-Wide Web server running on the Linux operating system. We demonstrate the utility of Webmonitor by measuring and understanding the performance of a Pentium-based PC acting as a dedicated WWW server. Our workload uses a file size distribution with a heavy tail. This captures the fact that Web servers must concurrently handle some requests for large audio and video files, and a large number of requests for small documents, containing text or images. Our results show that in a Web server saturated by client requests, over 90% of the time spent handling HTTP requests is spent in the kernel. Furthermore, keeping TCP connections open, as required by TCP, causes a factor of 2-9 increase in the elapsed time required to service an HTTP request. Data gathered from Webmonitor provide insight into the causes of this performance penalty. Specifically, we observe a significant increase in resource consumption along three dimensions: the number of HTTP processes running at the same time, CPU utilization, and memory utilization. These results emphasize the important role of operating system and network protocol implementation in determining Web server performance.

Generating Representative Web Workloads for Network and Server Performance Evaluation

One role for workload generation is as a means for understanding how servers and networks respond to variation in load. This enables management and capacity planning based on current and projected usage. This paper applies a number of observations of Web server usage to create a realistic Web workload generation tool which mimics a set of real users accessing a server. The tool, called Surge (Scalable URL Reference Generator) generates references matching empirical measurements of 1) server file size distribution; 2) request size distribution; 3) relative file popularity; 4) embedded file references; 5) temporal locality of reference; and 6) idle periods of individual users. This paper reviews the essential elements required in the generation of a representative Web workload. It also addresses the technical challenges to satisfying this large set of simultaneous constraints on the properties of the reference stream, the solutions we adopted, and their associated accuracy. Finally, we present evidence that Surge exercises servers in a manner significantly different from other Web server benchmarks.

On the Interaction Between an Operating System and Web Server

This paper examines how and why web server performance changes as the workload at the server varies. We measure the performance of a PC acting as a standalone web server, running Apache on top of Linux. We use two important tools to understand what aspects of software architecture and implementation determine performance at the server. The first is a tool that we developed, called WebMonitor, which measures activity and resource consumption, both in the operating system and in the web server. The second is the kernel profiling facility distributed as part of Linux. We vary the workload at the server along two important dimensions: the number of clients concurrently accessing the server, and the size of the documents stored on the server. Our results quantify and show how more clients and larger files stress the web server and operating system in different and surprising ways. Our results also show the importance of fixed costs (i.e., opening and closing TCP connections, and updating the server log) in determining web server performance.

A Two-Tiered On-Line Server-Side Bandwidth Reservation Framework for the Real-Time Delivery of Multiple Video Streams

The advent of virtualization and cloud computing technologies necessitates the development of effective mechanisms for the estimation and reservation of resources needed by content providers to deliver large numbers of video-on-demand (VOD) streams through the cloud. Unfortunately, capacity planning for the QoS-constrained delivery of a large number of VOD streams is inherently difficult as VBR encoding schemes exhibit significant bandwidth variability. In this paper, we present a novel resource management scheme to make such allocation decisions using a mixture of per-stream reservations and an aggregate reservation, shared across all streams to accommodate peak demands. The shared reservation provides capacity slack that enables statistical multiplexing of peak rates, while assuring analytically bounded frame-drop probabilities, which can be adjusted by trading off buffer space (and consequently delay) and bandwidth. Our two-tiered bandwidth allocation scheme enables the delivery of any set of streams with less bandwidth (or equivalently with higher link utilization) than state-of-the-art deterministic smoothing approaches. The algorithm underlying our proposed frame-work uses three per-stream parameters and is linear in the number of servers, making it particularly well suited for use in an on-line setting. We present results from extensive trace-driven simulations, which confirm the efficiency of our scheme especially for small buffer sizes and delay bounds, and which underscore the significant realizable bandwidth savings, typically yielding losses that are an order of magnitude or more below our analytically derived bounds.

Distributed Packet Rewriting and its Application to Scalable Server Architectures

To construct high performance Web servers, system builders are increasingly turning to distributed designs. An important challenge that arises in distributed Web servers is the need to direct incoming connections to individual hosts. Previous methods for connection routing have employed a centralized node which handles all incoming requests. In contrast, we propose a distributed approach, called Distributed Packet Rewriting (DPR), in which all hosts of the distributed system participate in connection routing. We argue that this approach promises better scalability and fault-tolerance than the centralized approach. We describe our implementation of four variants of DPR and compare their performance. We show that DPR provides performance comparable to centralized alternatives, measured in terms of throughput and delay under the SPECweb96 benchmark. Finally, we argue that DPR is particularly attractive both for small scale systems and for systems following the emerging trend toward increasingly intelligent I/O subsystems.

The Cyclone Server Architecture: Streamlining Delivery of Popular Content

We propose a new technique for efficiently delivering popular content from information repositories with bounded file caches. Our strategy relies on the use of fast erasure codes (a.k.a. forward error correcting codes) to generate encodings of popular files, of which only a small sliding window is cached at any time instant, even to satisfy an unbounded number of asynchronous requests for the file. Our approach capitalizes on concurrency to maximize sharing of state across different request threads while minimizing cache memory utilization. Additional reduction in resource requirements arises from providing for a lightweight version of the network stack. In this paper, we describe the design and implementation of our Cyclone server as a Linux kernel subsystem.

Timeliness via Speculation for Real-Time Databases

Various concurrency control algorithms differ in the time when conflicts are detected, and in the way they are resolved. In that respect, the Pessimistic and Optimistic Concurrency Control (PCC and OCC) alternatives represent two extremes. PCC locking protocols detect conflicts as soon as they occur and resolve them using blocking. OCC protocols detect conflicts at transaction commit time and resolve them using rollbacks (restarts). For real-time databases, blockages and rollbacks are hazards that increase the likelihood of transactions missing their deadlines. We propose a Speculative Concurrency Control (SCC) technique that minimizes the impact of blockages and rollbacks. SCC relies on the use of added system resources to speculate on potential serialization orders and to ensure that if such serialization orders materialize, the hazards of blockages and roll-backs are minimized. We present a number of SCC-based algorithms that differ in the level of speculation they introduce, and the amount of system resources (mainly memory) they require. We show the performance gains (in terms of number of satisfied timing constraints) to be expected when a representative SCC algorithm (SCC-2S) is adopted.

Geometric Generalizations of the Power of Two Choices

A well-known paradigm for load balancing in distributed systems is the``power of two choices,''whereby an item is stored at the less loaded of two (or more) random alternative servers. We investigate the power of two choices in natural settings for distributed computing where items and servers reside in a geometric space and each item is associated with the server that is its nearest neighbor. This is in fact the backdrop for distributed hash tables such as Chord, where the geometric space is determined by clockwise distance on a one-dimensional ring. Theoretically, we consider the following load balancing problem. Suppose that servers are initially hashed uniformly at random to points in the space. Sequentially, each item then considers d candidate insertion points also chosen uniformly at random from the space,and selects the insertion point whose associated server has the least load. For the one-dimensional ring, and for Euclidean distance on the two-dimensional torus, we demonstrate that when n data items are hashed to n servers,the maximum load at any server is log log n / log d + O(1) with high probability. While our results match the well-known bounds in the standard setting in which each server is selected equiprobably, our applications do not have this feature, since the sizes of the nearest-neighbor regions around servers are non-uniform. Therefore, the novelty in our methods lies in developing appropriate tail bounds on the distribution of nearest-neighbor region sizes and in adapting previous arguments to this more general setting. In addition, we provide simulation results demonstrating the load balance that results as the system size scales into the millions.

dPAM: A Distributed Prefetching Protocol for Scalable Asynchronous Multicast in P2P Systems

We leverage the buffering capabilities of end-systems to achieve scalable, asynchronous delivery of streams in a peer-to-peer environment. Unlike existing cache-and-relay schemes, we propose a distributed prefetching protocol where peers prefetch and store portions of the streaming media ahead of their playout time, thus not only turning themselves to possible sources for other peers but their prefetched data can allow them to overcome the departure of their source-peer. This stands in sharp contrast to existing cache-and-relay schemes where the departure of the source-peer forces its peer children to go the original server, thus disrupting their service and increasing server and network load. Through mathematical analysis and simulations, we show the effectiveness of maintaining such asynchronous multicasts from several source-peers to other children peers, and the efficacy of prefetching in the face of peer departures. We confirm the scalability of our dPAM protocol as it is shown to significantly reduce server load.

Demand-based Document Dissemination for the World-Wide Web

We analyzed the logs of our departmental HTTP server http://cs-www.bu.edu as well as the logs of the more popular Rolling Stones HTTP server http://www.stones.com. These servers have very different purposes; the former caters primarily to local clients, whereas the latter caters exclusively to remote clients all over the world. In both cases, our analysis showed that remote HTTP accesses were confined to a very small subset of documents. Using a validated analytical model of server popularity and file access profiles, we show that by disseminating the most popular documents on servers (proxies) closer to the clients, network traffic could be reduced considerably, while server loads are balanced. We argue that this process could be generalized so as to provide for an automated demand-based duplication of documents. We believe that such server-based information dissemination protocols will be more effective at reducing both network bandwidth and document retrieval times than client-based caching protocols [2].

AIDA-Based Real-Time Fault-Tolerant Broadcast Disks

The proliferation of mobile computers and wireless networks requires the design of future distributed real-time applications to recognize and deal with the significant asymmetry between downstream and upstream communication capacities, and the significant disparity between server and client storage capacities. Recent research work proposed the use of Broadcast Disks as a scalable mechanism to deal with this problem. In this paper, we propose a new broadcast disks protocol, based on our Adaptive Information Dispersal Algorithm (AIDA). Our protocol is different from previous broadcast disks protocols in that it improves communication timeliness, fault-tolerance, and security, while allowing for a finer control of multiplexing of prioritized data (broadcast frequencies). We start with a general introduction of broadcast disks. Next, we propose broadcast disk organizations that are suitable for real-time applications. Next, we present AIDA and show its fault-tolerance and security properties. We conclude the paper with the description and analysis of AIDA-based broadcast disks organizations that achieve both timeliness and fault-tolerance, while preserving downstream communication capacity.