Performance Evaluation of Multicast Behavior in Congested Networks

Sammanfattning: Compuverde’s software-defined storage product uses multicast for the communication between servers in a cluster. The product makes use of IP UDP multicast for sending status messages between the servers that forms the storage cluster. The storage clusters capacity and performance scales linear to the number of servers in the cluster. The problem is that the multicast traffic also increases with the number of nodes. All nodes send to all other nodes in the cluster. In this document, we present a proposal on evaluation of IP multicast behavior in a network congested with traffic similar to that produced by Compuverde’s product. IP multicast is a method of sending Internet Protocol (IP) datagrams to a group of interested receivers in a single transmission. In order to provide an efficient, timely, and global many-to-many distribution of data, and as such may become the broadcast medium of choice in the future, IP multicasting is used[1]. The main benefit of IP Multicast is that it reduces the bandwidth consumption when data from a sender must reach multiple receivers. We are interested in studying the effects on the network when we send multicast packets at a rate closed to the operational limit of the switch. To be able to study this behavior at larger scale Compuverde’s will provide a cluster with 48 servers all connected to the same switch. In addition, we will compare the behavior of IPv4 multicast traffic to that of IPv6. Aims and Objectives: Our aim of my thesis is mainly to focus on IP multicast and compare the IPv4 multicast performance results to the results from IPv6 multicast. In addition, a C++ tool for generating multicast traffic will be developed on Linux. A detailed study on IP multicast (IPv4, IPv6). Detailed study on the design and efficient implementation of a multicast traffic generating tool. Detailed study on the switch that will be used in the project. Additional switches may be provided by BTH. Detailed study on the pattern of dropped packets when traffic rate approaches operational limit and other related impairments on QoS metrics (e.g., CPU utilization). Methods: The method is to develop a tool that will generate multicast load towards servers in a cluster. The data sent as multicast packets shall consist of information that will make it possible to detect packet loss on the receiving servers if the network gets congested. The first version of the tool shall use existing socket classes that are based on the IPv4 protocol and shall be written in C++. The tool shall be able to run in two modes at the same time: client mode and server mode. The server part of the tool shall subscribe to a predefined multicast address and receive incoming multicast packages. The client part of the tool shall send data packages to the same predefined multicast address at a configurable rate that will increase over time. The data in the packet that will be sent shall be constructed in a way that lets the receiver (server) detect if a packet is lost in transmission. The load should start small with a small number of servers in the cluster, and then in steps scale up the number of servers, until a maximum of 48 servers is reached. The rate that the multicast packets is sent should also be increased, until the switch gets overloaded and starts to drop packets. The pattern of how packets are dropped should be observed. For example such as, if it is biggerlarge chunks of packets that get dropped or if it is every second packet that gets dropped. The second version of the tool shall support IPv6 multicast. The second round of tests should be performed in a way that makes them comparable to the results from the IPv4 tests so it is possible to draw conclusions if one protocol performs better or is more reliable. Result: The maximum number of IPv4 packets a switch can handle is 140 packets per second. The maximum number of IPv6 packets a switch can handle is 6 packets per second. The CPU utilization is more while multicasting the IPv4 packets than while multicasting IPv6 packets by using switch, 95 Nodes. Conclusion: The IPv4 is most efficient protocol than IPv6 protocol while sending the packets at very high data rate. The CPU utilization is more higher for sending with the IPv6 protocol packets than with the IPv4 protocol.