The production network at Arizona State University uses a commercial load balancer for \texttt{syslog} messages from the firewalls at the data center perimeter. The logs are carried in UDP packets and the load balancer divides the load among servers based on the source IP address. This may cause a server to be overwhelmed due to the high incoming data rate, and create log files unequal in size. These logs are processed by \texttt{splunk}, an application for analyzing the massive streams of data generated by IT systems. In a testbed with an OpenFlow switch and three \texttt{rsyslog} servers, fed by a mirror of the firewall \texttt{syslog} traffic, we study controllers implementing round-robin, random, and load-based load balancing policies. All policies achieve delivery ratios comparable with the commercial load balancer. When servers fail or go down for maintenance, the load-based policy produces log files the most balanced in size. The results suggest that an OpenFlow load balancer has performance competitive with a commercial load balancer, and can equalize the log file size at servers which improves the efficiency of \texttt{splunk}.
OpenFlow versus Commercial Load Balancers in a Campus Network
01.09.2015
212500 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
New Design Method for Spring Balancers
Online Contents | 2001
|Dynamic Openflow-Controlled Optical Packet Switching Network
British Library Online Contents | 2015
|