Plug-n-Serve: Load Balancing Web Traffic using OpenFlow

sigcomm’09 demo conference site copy 2015/10/03 16:23:48, cited 110+

Abstract Effective load-balancing systems for services hosted in unstructured networks need to take into account both the congestion of the network and the load on the servers. In this demonstration, we illustrate a comprehensive load-balancing solution that works well for such networks. The system we showcase, called Plug-n-serve, tries to minimizes response time by controlling the load on the network and the servers using customized flow routing. The demonstration shows how the Plug-n-Serve system works within our deployment in the CS building at Stanford University. Besides the base behavior, we show the effect of dynamically adding and removing computing resources to the system, increasing the request arrival rate, altering the CPU or network load of each request, and the changing load-balancing algorithms.

large web sites balance load over many HTTP severs, and there exist commercial products: Foundry ServerIron Load Balancer, Microsoft’s Network Load Balancing … Load balancing may be oblivious (e.g., spreading the requests equally over all servers, without regard for their load), or stateful (e.g., sending requests to the least-loaded server).

data-center or a dedicated web-hosting service, the HTTP servers are connected by a regular, over-provisioned network; the load balancer usually does not consider the network state when load-balancing across servers.
unstructured network (enterprise, campus networks) that are not custom-built for running server farms: the substantial background traffic and the potential topological biases can significantly affect the performance of network-oblivious load balancing, and inflate the response time (duration from issuing the HTTP request to the complete receipt of the response).

use OpenFlow architecture to measure the state of the network, and to directly control the paths taken by the HTTP requests.