Americas

  • United States
by John Bass

How we did it

Reviews
May 05, 20033 mins
Computers and PeripheralsNetworking

How we tested the HP blade servers.

The e-Class and p-Class servers are not useful as heavy file servers because the local hard drive capacity is relatively low. With this in mind, these products are useful as static-content Web servers. To test this capability, we ran HTTP performance tests using a Spirent Communications WebAvalanche tester.

We loaded the server with various sizes of HTTP file requests to see how many files the blade could service in 1 second. These tests were repeated for both Windows 2000 Advanced Server running Internet Information Server (IIS) and for a Red Hat Linux 8.0 installation running Apache Web server.

The HP blade servers are not designed to act as large capacity database servers because of the limited processor capacity. But because the blade servers are well suited for Web service, it make sense that they should be able to handle Secure Sockets Layer (SSL) computations used for secure Web transactions. To create the SSL transaction load, we used Spirent’s WebAvalanche tester.

The WebAvalanche tester was configured to request a secure Web page loaded on the blade servers running Win 2000 Advanced Server with IIS. The test was repeated against server blades running Apache Web server on a Linux RedHat 8.0 installation. The file requested is a 51-byte file to minimize the amount of network utilization during the test. To maximize the load on the server’s processors, all session caching is disabled, so the SSL keys are calculated with each Web request.

The four Fast Ethernet ports of the WebAvalanche were connected to four ports of a Cisco 2948. The server blade chassis is connected to another port of the Cisco 2948. All Ethernet ports were configured for maximum rate, full duplex operation.

Because these benchmarking tests were designed to apply the greatest amount of stress possible to the server processors, the test results should in no way indicate the number of SSL transactions or users that the server can support in a real environment. It would be safe to say that the number of supportable “real” users likely would be much higher than our metrics.

A similar Web test was run against the blade servers without enabling SSL. We ran three test cases with 1K-byte, 28K-byte and 1M-byte file sizes.

Configuration for our blade server tests

OS Configuration
Windows:
Windows 2000 Advanced Server
IIS 5.0
Tool to monitor performance: Performance Monitor
Linux:
Red Hat 8.0
Apache 2.0.40
Tux 2.2.7
Tool to monitor performance: Top
SSL Web Test Configuration
On Windows:
1. Install IIS on the blade
2. Install the Certificate Authority (CA)
3. Create a Server Certificate for the default Web site
a.Use IIS Server Certificate wizard
b.Issue the Server certificate to the CA
c.Make the CA approve the certificate
4.Enable SSL for the default Web site
5.Create a default html file called “Default.html” (~1KB)
On Linux:
1.Set the following parameters in the “httpd.conf” file
SSLEngine on (This turns on the SSL engine on this host)
SSLCipherSuite:ALL (This enables all SSL ciphers available)
2.Create a 1Kb file called “index.html” in the /var/www/html DocumentRoot directory

Back to main review: “HP blade servers”