Transcript Lecture 23
CS 453 Computer Networks Lecture 23 Network Management and Performance Performance Issues Clearly, users of networks expect them to perform well… …and usually notice when they do not… … and expect the network management to fix the problems. Performance Issues Down routes – network routes fail or shutdown for a variety of reasons, network management software must be able to detect route or link failures, … …and provide a means to resolve or work around the problem…. For example, rerouting traffic, until preferred routes are restored Performance Issues Congestion – Networks (particularly WANs) have multiple routes to provide end-to-end connectivity There are usually one or more preferred routes… But these preferred routes often aggregate traffic from multiple sources… Which can overload links and create congestion (compounding problem) Network management systems should be able to— Recognize congestion …and shed traffic to other routes, if possible Performance Issues Low end processing hardware/software High performance networks connected to old/slow hardware (routers, PCs) can cause performance problems Messages arrive faster than the hardware can process them… Messages are lost, and retransmitted,… And this problem can compound if the overload is not transient Performance Issues Synchronous overloads – Suppose a bad message is broadcast to a network of 10,000 machines…. …and each of these machines replies with an error message… Result can be massive overload of the network Broadcast Storm UDP protocol changes to inhibit error reports for broadcast messages Performance Issues Synchronous Overloads Imagine the simultaneous restart of all machines/routers in a network… … like a regional power failure Every machine on the network will start and may issue a DHCP Discovery request, simultaneously… Routers will update their routes and configurations, Network hosted software will reload And the network will come to it’s knees Performance Issues Synchronous Overloads This is not far fetched Happens all of the time on a modest scale Recently, happened on a global scale… Due to a MS Windows automatic update Update required a restart Something like 6 million computers worldwide restarted at the same time Performance Issues Tuning Insufficient buffer space in routers, source/destination computers Scheduling algorithm giving too little priority to processing incoming data, drops messages, causes retransmits Timer settings Too short – messages in-flight reported as lost Too long – waste time waiting for lost messages Performance Issues Inefficient use of resources For example, high performance network with low feed Performance Issues Jitter We want a short mean arrival time Also want small standard deviation for arrival time Large standard deviation in arrival time is called Jitter Jitter big problem with new services like VoIP Streaming audio Streaming video Can mitigate some Network Management More things to see http://www.cisco.com/univercd/cc/td/doc/cisintwk/ito_doc/snmp.htm http://en.wikipedia.org/wiki/Network_performance_management http://en.wikipedia.org/wiki/Snmp http://www.cotse.com/tools/netman.htm TCP/IP http://www.redbooks.ibm.com/Redbooks.nsf/RedbookAbstracts/gg243376.html?Open