Compaq ProLiant 1000 Performance Analysis and Tuning of Raptor's Eagle NT 3.06 - Page 4

278A/0497 WHITE PAPER (cont.) ... variability of multi-protocol loads through gateways/firewalls. Multi-protocol benchmarks allow firewalls to be stressed in ways, which closely simulate real network traffic. Of the four benchmarks described below, NSTL's Benchmark enables two protocols to be used, HTTP and FTP, which are the two most used on the Internet and thus the tool used to test performance of the firewall in this paper. WebStone Webstone measures the raw throughput of a standard HTTP workload. It is measured across two main variables--latency, in seconds, and the number of connections per second. Other data can also be collected, such as throughput in bits per second. The benchmark itself uses a client/server architecture, and each client runs a configuration file that tells it which server to connect to, how long to maintain the connection, and which URLs to fetch. WebBench WebBench reports two types of results: the average number of requests per second the HTTP server handled, and the average number of bytes per second that moved between the clients and the server. The controller computes these results by collecting each client's results as that client finishes the test. SpecWeb SpecWeb measures the response time for server requests across a number of different workloads. It sends HTTP requests to the server, based on defined workload parameters, and calculates the overall throughput at the end of a run. The workload information that the benchmark takes into account includes request rate and request type, file set, database transactions, security, and slow networks. NSTL Benchmark This benchmark was designed specifically to stresses the ability of the firewall to route traffic based on a set of rules. It simulates real usage of the firewall by directing heavy loads of FTP and HTTP traffic through the firewall. The toolkit also allows WAIS traffic and CGI requests to be sent. The results are measured in overall transaction time for each client data transfer for a set of virtual clients. This is then broken down into TPM (Transactions per Minute) based on the number of URLs and FTP files requested. NSTL METHODOLOGY OF INTERNET FIREWALLS This section describes in detail the NSTL methodology used for the tests and the alterations made to the methodology to allow more traffic to pass through the firewall. Configuration The NSTL methodology was designed to stress the ability of the firewall to route traffic based on a set of rules. Heavy loads of HTTP and FTP traffic are generated for requests through the firewall. The overall transaction time for each client transfer is measured and reported under various load conditions. The original methodology used three scenarios for measuring performance through the firewall. Together, the three scenarios did not provide an effective challenge to the firewall, so changes were made to NSTL's benchmark configuration to allow more traffic to pass through the firewall from both sides. 4