Kemp Technologies' LM 1500

I had an LM up and running in minutes in our Green Bay, Wis., Real-World Labs®. I found the straightforward user interface easy to navigate, but some admins might find the device's use of Linux/BSD terminology to refer to its Ethernet ports (ethX) and the device's labeling (LANx) confusing.

After installing the device, I configured a virtual service and added three Web servers emulated by a Spirent WebReflector to the service. The LM, which uses configurable health-checking (HTTP, ICMP, TCP and so on) to determine whether real servers are available, immediately discovered the Web servers, while marking the virtual service as available.

Using a Spirent WebAvalanche to simulate no more than 4,600 connections per second (the LM claims a maximum of 4,000), I configured the WebAvalanche to emulate clients requesting a page comprised of seven files. The LM provided simple system statistics including CPU utilization, available memory, network bandwidth and connections per second to both virtual service and real servers.

The LM performed admirably in an L4 configuration, serving up an average 2,289 transactions per second with barely noticeable latency (less than 2 ms). The device held its CPU utilization between 60 percent and 70 percent until network bandwidth went past 90 Mbps, which caused CPU rates to climb to 99 percent as the amount of traffic oversubscribed its 100-Mbps link to my Cisco Catalyst 6500.

Follow the Rules

The LM offers three options for creating content-switching rules: postfix, prefix and regular expressions. Rules are created separately from virtual services, so they can be used by more than one virtual service. I configured two rules, one postfix and one regular expression, which looked for requests for GIF images. I checked the Enable button next to Content Switching in my virtual service and assigned rules to individual real servers. The system automatically configures a Default rule, so I could assign my postfix rule to two real servers and set the third as the default, to service all requests for non-GIF content.

I ran the same client requests in an L7 configuration, and the connections to real servers appeared to correlate 1:1 with client connections. The vendor verified the LM doesn't do TCP multiplexing (reuse TCP connections) on the back end.I reconfigured the device to use the regular expression rule and reran the test. As expected, the device performed slightly fewer transactions per second and showed higher latency in L7 configurations, with the postfix rule set performing slightly higher than its regular expression equivalent. SMBs are likely to use postfix or prefix, though simple regular expressions are a good option for department-level and straightforward enterprise load balancing needs.

SSL acceleration is a standard option on the LM 1500. Performance should be adequate for most simple L4/L7 load balancing needs, at a price that's easy on the wallet.

Lori MacVittie is a Network Computing senior technology editor working in our Green Bay, Wis., labs. She has been a software developer, a network administrator, and a member of the technical architecture team for a global transportation and logistics organization. Write to her at [email protected].