Performance Analysis of NIC Caching for Web Server Clusters on System Area Network
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
High performance Web servers have become a mandate for an increasing number of network-based applications and services. To meet this need, System Area Networks (SANs) have been designed to free up valuable server resources such as CPU cycles, to improve performance of applications running on the server. In this paper, we examine the architecture of Web Server Clusters based on SANs and exploit the abundant local memory available in programmable Network Interface Cards (NICs) for SANs. We explore two NIC caching schemes, namely, exclusive caching and inclusive caching, to reduce disk accesses and communication cost (in terms of data transfer latency), respectively. In addition, we explore impact of an I/O interconnection architecture, cache replacement policy and scalability on performance with the NIC caching schemes in Web Server Clusters. To maximize performance benefit from those caching schemes, we analyze performance implication of NIC caching on the Web server under several Web workloads of different characteristics. Moreover, we developed a simulator which is validated with an 8-node prototype implementation for performance evaluation. We conduct extensive experiments to compare the performance of Web servers with or without using NIC caching. Our results show that the proposed NIC caching schemes significantly increase throughput over the original system which does not use the NIC caching. The exclusive caching scheme becomes beneficial when a workload shows a low popularity skewness, while the inclusive caching scheme is particularly effective with workloads with the large file sizes and high popularity skewness. The results provide important guidance for tuning NIC caching schemes in Web Server Clusters.
Abstract
High performance Web servers have become a mandate for an increasing number of network-based applications and services. To meet this need, System Area Networks (SANs) have been designed to free up valuable server resources such as CPU cycles, to improve performance of applications running on the server. In this paper, we examine the architecture of Web Server Clusters based on SANs and exploit the abundant local memory available in programmable Network Interface Cards (NICs) for SANs. We explore two NIC caching schemes, namely, exclusive caching and inclusive caching, to reduce disk accesses and communication cost (in terms of data transfer latency), respectively. In addition, we explore impact of an I/O interconnection architecture, cache replacement policy and scalability on performance with the NIC caching schemes in Web Server Clusters. To maximize performance benefit from those caching schemes, we analyze performance implication of NIC caching on the Web server under several Web workloads of different characteristics. Moreover, we developed a simulator which is validated with an 8-node prototype implementation for performance evaluation. We conduct extensive experiments to compare the performance of Web servers with or without using NIC caching. Our results show that the proposed NIC caching schemes significantly increase throughput over the original system which does not use the NIC caching. The exclusive caching scheme becomes beneficial when a workload shows a low popularity skewness, while the inclusive caching scheme is particularly effective with workloads with the large file sizes and high popularity skewness. The results provide important guidance for tuning NIC caching schemes in Web Server Clusters.
참고문헌 (36)
- V. Almeida, "Characterizing Reference Locality in the WWW", pp.92~103 (1996), In Proceeding of the Fourth International Conference on Parallel and Distributed Information System (PDIS)
- "Workload Characterization of the 1998 World Cup Web Site"[URL]
- M. F. Arlitt, "Web Server Workload Characterization: The Search for Invariants", ACM SIGMETRICS Performance Evaluation Review, 24(1), pp.126~137 (1996)
- M. F. Arlitt, "Internet Web Servers: Workload Characterization and Performance Implicationd", IEEE, 1997, pp.5(5), 631~645
- M. Aron, "Efficient Support for P-HTTP in Cluster-Based Web Servers", pp.185~198 (1999), Proceedings of the USENIX 1999 Annual Technical Conference
- M. A. Blumrich, "Virtual Memory Mapped Network Interface for the SHRIMP Multicomputer", pp.142~153 (1994), Proceedings of the 21th annual International Symposium on Computer Architecture (ISCA)
- L. Breslau, "Web Caching and Zipf-like Distributions: Evidence and Implications", ~ (1999)
- P. Buonadonna, "Queue Pair IP: A Hybrid Architecture for System Area Networks", pp.247~256 (2002), Proceedings of the 29th annual International Symposium on Computer Architecture (ISCA)
- P. Buonadonna, "An Implementation and Analysis of the Virtual Interface Architecture", pp.1998, Proceedings of the ACM
- Cost-Aware WWW Proxy Caching Algorithms, "P. Cao", In Proceedings of the 1997 Usenix Symposium on Internet Technologies and Systems, ~, pp.USITS-97 (1997)
- V. Cardellini, "The State of the Art in Locally Distributed Web-Server Systems", ACM Computing Surveys (CSUR), 34(2), pp.236~311 (2002)
- User-Level Communication in Cluster-Based Servers, "E. V. Carrera", Proceedings of the Eighth International Symposium on High-Performance Computer Architecture, 248~259, pp.HPCA 02 (2002)
- A. A. Chien, "Design Challenges for High-Performance Network Interfaces, Guest Editor s Introduction", IEEE Computer, 31(11), pp.42~44 (1998)
- G. S. Choi, "Exploiting nic memory for improving cluster- based webserver performance", IEEE International Conference on Cluster Computing (2005)
- "Intel and Microsoft Corporations. Virtual Interface Architecture Specification. Version 1.0"[URL]
- Characteristics of WWW Client-based Traces, "C. Cunha", Boston University, ~, pp.Technical Report BU-CS-95-010 (1995)
- W. J. Dally, "Principles and Practices of Interconnenction Networks" (2004), Morgan Kaufmann
- T. V. Eicken, "U-Net: A User-level Network Interface of Parallel and Distributed Computing", pp.40~53 (1995), Proceedings of the 15th ACM Symposium on Operating Systems Principles (SOSP),
- T. V. Eicken, "Active Messages: A Mechanism for Integrated Communication and Computation", pp.256~266 (1992), Proceedings of the 19th Annual International Symposium on Computer Architecture
- R. Fonseca, "On the Intrinsic Locality Properties of Web Reference Streams", Proceedings of IEEE Infocom, pp.448~458 (2003)
- P. Geoffray, "OPIOM: Off-Processor I", O with Myrinet, ACM International Symposium on Cluster Computing and the Grid, pp.2001[DOI]
- "InfiniBand Architecture Specification, Vol. 1 & 2, Release 1.2"[URL]
- V. Karamcheti, "Software Overhead in Messaging Layers: Where Does the Time Go?", pp.51~60 (1994), Proc. Sixth Intl. Conf. on Architectural Support for Programming Languages and Operating Systems (ASPLOS-VI)
- "Adamic. Zipf, Power-laws, and Pareto a ranking tutorial. Available from"[URL]
- "Available from"[URL]
- S. Pakin, "High Performance Messaging on Workstations: Illinois Fast Messages (FM) for Myrinet", pp.1995, Proceedings of the 1995 ACM
- "PCIExpressTM Specifications"[URL]
- "Pennsylvania State University"[URL]
- "Pennsylvania State University Computer Science & Engineering"[URL]
- "Quadrics Ltd. QsNet HIGH PERFORMANCE INTERCONNECT. Available from"[URL]
- "Seagate Technology"[URL]
- J. Sedayao, "Proc. 2nd World Wide Web Conf: Mosaic and Web", ~ (1994)
- "UC Berkeley Home IP HTTP Traces"[URL]
- "Inc. Voltaire HCA 400LP InfiniBand 4X Host Channel Adapter. Available from"[URL]
- Can NIC memory in InfiniBand Benefit Communication Performance? A Study with Mellanox Adapter, "J. Wu", Computer and Information Science, The Ohio State University, ~, pp.Technical Report OSU-CISRC-4 (2004), 04- TR20
- H. Youb Kim, "Increasing Web Server Throughput with Network Interface Data Caching", pp.239~250 (2002), In Proceedings of the 10th International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS)