Unlike other Web load-balancing solutions, which begin as Layer 3 and Layer 4 switches, the CSS-100 was built from the ground up with Layer 4 and Layer 5 switching and content load-balancing. It offers a 5-Gbps ATM switching fabric for high-speed, low-latency switching. Twelve 100BASE-TX interfaces and one 100BASE-TX management port are available. The ATM backplane is transparent to the end user, and provides a fast, scalable switching core.

Directing Heavy Traffic The CSS-100 is a phenomenal tool for balancing Internet traffic on your LAN. At the core of the CSS-100 is a wire-speed Layer 3 IP switch. You can assign each port on the CSS-100 to a unique VLAN (virtual LAN), or you can assign multiple ports to a single VLAN with Layer 2 bridging in between. In the lab, I configured the CSS-100 with three VLANs (each containing four ports) and then assigned an IP interface to each.

The CSS-100 provides NAT (Network Address Translation) for hosts on the back end of your server farm. I configured VLAN No. 1 to use public IP addressing and attached a Netscape Communications Corp. Communicator client to that network. VLAN No. 2 and VLAN No. 3 were configured to use private network addressing. I also attached Web servers to both networks.

The CSS-100 provides multiple methods to verify server resource availability. After I defined the VLANs, I configured the CSS-100 to be aware of each server resource--called a service --on the network. Connectivity to each service may be monitored either by HTTP "GETting" a page on the Web server or by ICMP (Internet Control Message Protocol) "PINGing" the Web server itself. By monitoring HTTP connectivity, the switch ensures that the hardware, software and TCP stack on the server are all functioning properly. In contrast, monitoring ICMP only verifies the hardware and the TCP stack. I defined seven services--one for each server on the switch--using the HTTP keep-alive monitoring option to ensure server connectivity.

Once I defined these services, I grouped them into load-sharing "virtual servers." In this process, an "owner" host site is defined with a virtual IP address. Each TCP stream is balanced round-robin across the servers in the group.

The CSS-100 enhances the load-balancing process by balancing based on response time. This unique algorithm, called ArrowPoint Content Aware (ACA) load-balancing, dynamically monitors server response times and attempts to adjust traffic flows to best suit each server's capacity. By monitoring server response times, servers with more capacity will receive more traffic. Compared to round-robin load-balancing, the traffic is equally distributed among all the servers without consideration of server load.

Another unique feature of the CSS-100 is its ability to switch based on URL content. To test this feature, I created two distinct groups of content on my Web servers--deliberately splitting the servers in VLAN No. 2 so that half of them included a certain type of unique content, while the other half lacked that content. I configured a rule such that all content matching the "music" pattern would be directed to servers with that unique content. All other traffic would default to the remaining Web servers. I tested each server directly to verify that that content did not exist, and then addressed the virtual IP address of the CSS-100. Clients requesting content from the "music" Web page were properly directed to the content-specific servers.