Latest News

Performance Testing Interview with Cyberheat System Admin

06/28/02 -- Performance test results for an 8-node load-balanced cluster prove the powerful impact of ProxyPass. Greatly decreased load averages across all cluster nodes translates into a direct $$ savings and maintenance reduction for Cyberheat, Inc., who have kindly agreed to publish their test findings.

In May 2002, Cyberheat began an intensive performance evaluation of the ProxyPass on an 8-node load-balanced cluster used for part of their "Topcash" program. Four "control" servers continued to run iProtect while the remaining four machines ran ProxyPass. The ProxyPass machines showed an immediate advantage over their counterparts by quickly propagating block information to the other cluster nodes. This mechanism for immediately sharing block information amongst cluster nodes is tremendously important in load-balanced setups. Without it, web sites experience "partial-block": a block will exist on one cluster node but not on another. The most important feature of ProxyPass proved to be its ability to immediately deny abused proxies.

Tim, lead system administrator for the Topcash program, provided key answers to our interview questions:

ProxyPass Q: Does ProxyPass, as it currently stands, meet your requirements for protection against password cracking and password sharing violations?
CyberHeat A: It does a good job of spotting the attacks and blocking proxies.

ProxyPass Q: Are you satisfied with the feature set? Administration interface? Ease of use for installation and maintenance?
CyberHeat A: I am.

ProxyPass Q: Is the performance of your servers adversely impacted at all by the ProxyPass?
CyberHeat A: Just the opposite. Under a large attack the ProxyPass servers have a load that is approx. 20% of the other servers.

ProxyPass Q: Based on your experience, does the ProxyPass outperform previous solutions (i.e. iProtect, Pennywize)?
CyberHeat A: Yep. The lower system load combined with the ability to share info [throughout a cluster] does it. These [numbers below]are basic system load averages (1 min.). The only difference that I'm aware of between the system groups are that 15, 17, 19, and 20 are running iProt. 21->24 are running ProxyPass. All the servers are dual PIII (Coppermine) 1GHz processors. All of them have 1GB memory except for kw20 and kw23. Those have 2GB. All are running Linux. It's amazing but servers running ProxyPass are showing a load average that is 5 - 20 times less than the servers running iProtect. This sample was taken during a heavy brute force attack.

Servers -without- ProxyPass

Servers -with- ProxyPass