计算机系统/网络性能评价 2008

计算机系统/网络性能评价 性能评价的应用： 研究，设计，运行，改进 系统，硬件，软件 ，协议，算法－ 计算机，网络，其他 改进系统的目的 不增加系统配置，但能提高原有系统性能 不增加系统配置，但能使系统现有负载能力增加 改善系统配置，提高系统性能 改进计算机系统性能 的过程 定义性能指标 建立系统模型（测量，仿真，解析）及负载模型 评价性能 （测量，仿真，解析） 重复以上步骤

性能指标 系统的处理能力 吞吐率是指系统在单位时间量，所处理的事务量，作业个数等等 系统的响应能力 响应时间：这主要是指系统在交互工作方式下，对终端用户的响应能力，响时间可以定义为：从用户在终端键盘上输入最后一个字符的时刻到终端输出第一个字符的时刻这一时间区间 周转时间：主要用在批处理方式，它表示从作业由输入设备提交给系统到结果在输出设备上出现的这一段时间 排队时间：它表示作业在队列中的等待时间和系统对作业的服务时间两部分的总和。 利用率 在评价者所考察的一段时间Ｔ内，系统的某一部分（硬件或软件）被使用的时间ｔ与Ｔ的比值称为系统这一部分的利用率

Network Performance Characteristics - Performance Metrics Throughput Latency - Delay Response time Loss Utilization Arrival rate, Departure rate Bandwidth, Capacity Routing （hop） Reliability

性能评价技术: 实验-测量，解析，仿真/模拟 性能评价技术是指对一个指定的计算机系统进行性能评价时所采用的方法，通过这种方法求出我们所期望的各项指标。 实验/测量 (measurement) 技术：通过测量设备或测量程序（软件）直接测量计算机系统的各种性能指标，或与之相关的量，然后由它们经过运算求出相应的性能的指标。 模型/建模 (modeling) 技术：对评价的计算机系统建立一个适当的模型，然后求出模型的性能指标，以便对计算机系统进行评价，该技术又分为解析技术和仿真技术两种。 解析 (analysis) 技术是采用数学分析方法，通过对系统的简化及解析模型的建立，以求得系统的性能。 仿真 (simulation) 技术是采用软件仿真原理，通过构造仿真模型，详尽、逼真地描述计算机系统。当模型按照系统本身的方式运行时，对系统的动态行为进行统计，从而得到有关的性能指标。 测量、解析、仿真技术是目前采用的主要性能评价技术，三者之间相互联系，相互验证，各有优缺点。 模拟（Emulation）---- simulation + experiment

Measurement Overview Measurement is necessary for understanding current system behavior and how new systems will behave How, when, where, what do we measure? Size, complexity and diversity of the Internet makes it very difficult to understand cause-effect relationships Measurement is meaningless without careful analysis Analysis of data gathered from networks is quite different from work done in other disciplines Measurement enables models to be built which can be used to effectively develop and evaluate new techniques Statistical models Queuing models Simulation models

Measurement Introduction Internet measurement is done to either analyze/characterize network phenomena or to test new tools, protocols, systems, etc. Measuring Internet performance is easier said than done What does “performance” mean? Workload (what and where you’re measuring) selection is critical Reproducibility is often essential Many tools have been developed to measure/monitor general characteristics of network performance ping and traceroute are two of the most popular These are examples of Active measurement tools Passive tools are the other major category Representative and reproducible workload generation will be a focus

Active Measurement Tools Send probe packet(s) into the network and measure a response Ping: RTT and loss Zing: one way Poisson probes Traceroute: path and RTT Nettimer: latest bottleneck bandwidth using packet pair method Pathchar: per-hop bandwidth, latency, loss measurement Pchar, clink: open-source reimplementation of pathchar Problem: measurement timescales vary widely T1 T0 Size/BW Tn+1 Tn Tn+1 - Tn = max(S/BW, T1 – T0)

Active Measurement: Ping Adding traffic for purposes of measurement Send probe packet(s) into the network and measure a response Trade-offs between accuracy and overhead Need careful methods to avoid introducing bias Ping: RTT and connectivity Host sends an ICMP ECHO packet to a target … and captures the ICMP ECHO REPLY Useful for checking connectivity, and RTT Only requires control of one of the two end-points Problems with ping Round-trip rather than one-way delays Some hosts might not respond

Active Measurement: Traceroute Traceroute: path and RTT TTL (Time-To-Live) field in IP packet header Source sends a packet with a TTL of n Each router along the path decrements the TTL “TTL exceeded” sent when TTL reaches 0 Traceroute tool exploits this TTL behavior Send packets with increasing TTL values source destination TTL=1 Time exceeded TTL=2 Send packets with TTL=1, 2, 3, … and record source of “time exceeded” message

Active Measurement: Pathchar for Links ---- per-hop capacity, latency, loss rtt(i+1) -rtt(i) Three delay components:  min. RTT (L) slope=1/c d How to infer d,c? L

Passive Measurement Tools Passive tools: Capture data as it passes by Logging at application level Packet capture software (tcpdump) uses packet capture filter (bpf,libpcap) Requires access to the wire Can have many problems (adds, deletes, reordering) Flow-based measurement tools SNMP tools Routing looking glass sites Problems LOTS of data! Privacy issues Getting packet scoped in backbone of the network

Passive Measurement Tools 利用以太网的广播特性,测量主机所在网段的业务流量 Tcpdump 测量 www.tju.edu.cn

Passive Measurement Tools 利用SNMP tools, 测量 TJU 出口业务流量 （24 小时）

Workload Generation Local and/or wide area experiments often require representative and reproducible workloads How do we select a workload? Currently HTTP makes up the majority of Internet traffic Trace-based workloads Capture traces and replay them Black-box method Synthetic workloads Abstraction of actual operation May not capture all aspects of workload Analytic workloads (model-based) Attempt to model workload precisely Very difficult

Analyzing Measured Data Analyzing measured data in networks is typically done using statistical methods Selecting appropriate analysis method(s) is critical Averaging Dispersion (variability) Correlations Regression analysis Distributional analysis Frequency analysis Principal-component analysis Cluster analysis Each form of analysis has strengths and weaknesses