Wenting Tang

MediSyn: a synthetic streaming media service workload generator

Currently, Internet hosting centers and content distribution networks leverage statistical multiplexing to meet the performance requirements of a number of competing hosted network services. Developing efficient resource allocation mechanisms for such services requires an understanding of both the short-term and long-term behavior of client access patterns to these competing services. At the same time, streaming media services are becoming increasingly popular, presenting new challenges for designers of shared hosting services. These new challenges result from fundamentally new characteristics of streaming media relative to traditional web objects, principally different client access patterns and significantly larger computational and bandwidth overhead associated with a streaming request. To understand the characteristics of these new workloads we use two long-term traces of streaming media services to develop MediSyn, a publicly available streaming media workload generator. In summary, this paper makes the following contributions: i) we model the long-term behavior of network services capturing the process of file introduction and changing file popularity, ii) we present a novel generalized Zipf-like distribution that captures recently-observed popularity of both web objects and streaming media not captured by existing Zipf-like distributions, and iii) we capture a number of characteristics unique to streaming media services, including file duration, encoding bit rate, session duration and non-stationary popularity of media accesses.

Measuring and characterizing end-to-end Internet service performance

Fundamental to the design of reliable, high-performance network services is an understanding of the performance characteristics of the service as perceived by the client population as a whole. Understanding and measuring such end-to-end service performance is a challenging task. Current techniques include periodic sampling of service characteristics from strategic locations in the network and instrumenting Web pages with code that reports client-perceived latency back to a performance server. Limitations to these approaches include potentially nonrepresentative access patterns in the first case and determining the location of a performance bottleneck in the second.This paper presents EtE monitor, a novel approach to measuring Web site performance. Our system passively collects packet traces from a server site to determine service performance characteristics. We introduce a two-pass heuristic and a statistical filtering mechanism to accurately reconstruct different client page accesses and to measure performance characteristics integrated across all client accesses. Relative to existing approaches, EtE monitor offers the following benefits: i) a latency breakdown between the network and server overhead of retrieving a Web page, ii) longitudinal information for all client accesses, not just the subset probed by a third party, iii) characteristics of accesses that are aborted by clients, iv) an understanding of the performance breakdown of accesses to dynamic, multitiered services, and v) quantification of the benefits of network and browser caches on server performance. Our initial implementation and performance analysis across three different commercial Web sites confirm the utility of our approach.

Modeling and generating realistic streaming media server workloads

Currently, Internet hosting centers and content distribution networks leverage statistical multiplexing to meet the performance requirements of a number of competing hosted network services. Developing efficient resource allocation mechanisms for such services requires an understanding of both the short-term and long-term behavior of client access patterns to these competing services. At the same time, streaming media services are becoming increasingly popular, presenting new challenges for designers of shared hosting services. These new challenges result from fundamentally new characteristics of streaming media relative to traditional web objects, principally different client access patterns and significantly larger computational and bandwidth overhead associated with a streaming request. To understand the characteristics of these new workloads we use two long-term traces of streaming media services to develop MediSyn, a publicly available streaming media workload generator. In summary, this paper makes the following contributions: (i) we propose a framework for modeling long-term behavior of network services by capturing the process of file introduction, non-stationary popularity of media accesses, file duration, encoding bit rate, and session duration. (ii) We propose a variety of practical models based on the study of the two workloads. (iii) We develop an open-source synthetic streaming service workload generator to demonstrate the capability of our framework to capture the models.

An SLA-oriented capacity planning tool for streaming media services

This paper addresses the problem of mapping the requirements of a known media service workload into the corresponding system resource requirements and accurately sizing a media server cluster to handle the workload. In this paper, we propose a new capacity planning framework for evaluating the resources needed for processing a given streaming media workload with specified performance requirements. The performance requirements are specified in a service level agreement (SLA) containing: i) basic capacity requirements that define the percentage of time the configuration is capable of processing the workload without performance degradation while satisfying bounds on system utilization; and ii) performability requirements that define the acceptable degradation of service performance during the remaining, non-compliant time and in case of node failures. Using a set of specially benchmarked media server configurations, the capacity planning tool matches the overall capacity requirements of the media service workload profile with the specified SLAs to identify the number of nodes necessary to support the required service performance.

Modular TCP Handoff Design in STREAMS-Based TCP/IP Implementation

Content-aware request distribution is a technique which takes into account the content of the request when distributing the requests in a web server cluster. A handoff protocol and TCP handoff mechanism were introduced to support content-aware request distribution in a client-transparent manner. Content-aware request distribution mechanisms enable the intelligent routing inside the cluster to provide the quality of service requirements for different types of content and to improve overall cluster performance.We propose a new modular TCP handoff design based on STREAMSbased TCP/IP implementation in HP-UX 11.0. We design the handoff functions as dynamically loadable modules. No changes are made to the existing TCP/IP code. The proposed plug-in module approach has the following advantages: flexibility-TCP handoff functions may be loaded and unloaded dynamically, without node function interruption; modularity- proposed design and implementation may be ported to other OSes with minimal effort.

Sizing the streaming media cluster solution for a given workload

The goal of the proposed benchmarking and capacity planning framework is to evaluate the amount of resources needed for processing a given workload while meeting the specified performance requirements. There are two essential components in our capacity planning framework: (i) the capacity measurements of different hardware and software solutions using a specially designed set of media benchmarks; and (ii) a media service workload profiler, called MediaProf, which extracts a set of quantitative and qualitative parameters characterizing the service demand. The capacity planning tool matches the requirements of the media service workload profile, SLA and configuration constraints to produce the best available cost/performance solution. In case of a multi-node configuration, the capacity planner performs a cluster sizing evaluation by taking into account the choice of load balancing solution.

Customized library of modules for STREAMS-based TCP/IP implementation to support content-aware request processing for Web applications

Content-aware request processing enables the intelligent routing and request processing inside a cluster to provide the QoS (quality of service) requirements for different types of content and to improve the overall cluster performance. A STREAMS-based TCP/IP implementation in HP-UX 11.0 provides a convenient framework to design a library of new STREAMS modules to support content-aware request distribution and differentiation inside a cluster. The proposed modules take into account the specifics of different cluster architectures and workload characteristics. These modules are designed as dynamically loadable modules, and no changes are made to the existing TCP/IP code. The proposed design has the following advantages: flexibility (new modules may be loaded and unloaded dynamically, without node function interruption) and modularity (the proposed modules may be ported to other operating systems with minimal effort; more importantly, the proposed STREAMS modules can be easily integrated and deployed into commercial operating systems, so that end-users may take advantage of these solutions much sooner).

Intelligent browser initiated server pushing

This paper proposes enhanced browser initiated pushing (BIP) approaches to reduce the download latency for web pages and to improve web server resource utilization. In our BIP approaches, upon receiving a HTTP request, the server actively pushes embedded contents if the permission is given by the client. By means of the pushing mechanism, the HTML web page will be downloaded in one RTT if the embedded contents and the HTML web page are on the same server. The number of requests a web server receives may be reduced significantly and the percentage of HTML documents that may benefit from BIP is quite high. Using trace-driven simulation, this paper presents an evaluation of enhanced BIP approaches that vary in the manner that client cache information is conveyed to the server.

Supporting global replicated services by a routing-metric-aware DNS

We propose domain name service (DNS) server extensions to support global replicated services. The DNS server collects routing metrics from routers for cached replicated servers. Upon receiving a DNS request from a host in its domain, the DNS automatically returns the IP address of a server available from an IP address pool that has the best routing metrics preferred by the host. Possible extensions to routers are proposed to support routing information collection. These router extensions work independently of the specific routing protocol used by the routers. The route metric collection mechanism is also independent of DNS extensions and may be used for other purposes. Our approach makes it much easier for content providers to support global replicated services flexibly and transparently. The network latency perceived by users is reduced substantially because of network proximity. Simulation results show that our approach is much better than DNS round robin. Moreover, the overhead introduced by our extension is negligible.

MediaGuard: a model-based framework for building streaming media services

A number of technology and workload trends motivate us to consider the appropriate resource allocation mechanisms and policies for streaming media services in shared cluster environments. We present MediaGuard -- a model-based infrastructure for building streaming media services -- that can efficiently determine the fraction of server resources required to support a particular client request over its expected lifetime. The proposed solution is based on a unified cost function that uses a single value to reflect overall resource requirements such as the CPU, disk, memory, and bandwidth necessary to support a particular media stream based on its bit rate and whether it is likely to be served from memory or disk. We design a novel, segment-based memory model of a media server to efficiently determine in liner time whether a request will incur memory or disk access when given the history of previous accesses and the behavior of the servers main memory file buffer cache. Using the MediaGuard framework, we design a novel, more accurate admission control policy for streaming media servers that accounts for the impact of the servers main memory file buffer cache. Our evaluation shows that, relative to a pessimistic admission control policy that assumes that all content must be served from disk, MediaGuard delivers a factor of two improvement in server throughput.