NOSes /

Wide-area load balancing

By Phil Hochmuth
Network World Linux in the Enterprise Newsletter, 02/21/01

Companies such as Inktomi and Akamai have done well by selling technologies that distribute Web content geographically, to connect users with Web servers that are closest to them. Users looking to implement similar technology themselves based on Linux might be interested in the developments of Supersparrow.

Supersparrow is an open-source software project aimed at creating a wide-area Web server load-balancing system based on Border Gateway Protocol (BGP). The software was developed by Simon Horman, a software engineer with VA Linux systems, and is licensed under the GNU Public License.

Supersparrow aims at improving the system of mirror sites used commonly to distribute Web downloads to the widest number of users possible. Instead of leaving it to users to manually pick and choose the best mirrors themselves, the software is supposed to automatically route a user to the optimal, or best-available, mirror site.

This is done using BGP, which is a commonly used routing protocol among Internet routers for determining along what path traffic will be sent. The software, when running on a Domain Name System (DNS) server or Web server, balances the load of incoming queries to identical, distributed Web servers. This is done by querying BGP-compatible routers (basically all Internet routers) to find the shortest paths for redirecting traffic.

While the concept is similar to Linux Virtual Server (LVS, which was examined in a recent newsletter), Supersparrow creator Horman claims Supersparrow is more adept for load-balancing servers across a WAN because it does not require a single point of contact for incoming traffic. (LVS requires traffic to pass through a single load-balancer box, which inspects packets at Layer 4 and forwards them to the appropriate destination.) All servers on the Supersparrow network are designed to bounce incoming traffic along the shortest path to the proper destination, which eliminates the bottleneck and potential single point of failure of the LVS' " load-balancer. "

Supersparrow is designed for installation on VA Linux servers running the VA-Enhanced Linux distribution, Version 6.2.3 or higher. It is also said to work with Red Hat Linux 6.2. The Supersparrow software also has hooks that can tie it together with Apache Web server. Supersparrow integrated with Apache is intended to provide more specified traffic redirection of Web content, rather than the more general redirection used in the DNS server implementation of Supersparrow.