Hey, Because each school's implementation is different, we load tested CAS and a few other applications. We loaded a non-people LDAP tree with a few thousand fake users, set CAS to fall back authenticate against that tree, and whipped up some test cases in Jmeter. In our experience we had more issues with the application that users were going to than CAS, but this sort of load test might be something you want to run.
Cheers Dave Wolowicz Manager of Web Services University of Victoria Systems [email protected] | (250) 721-6117 ________________________________________ From: Bryan E. Wooten [[email protected]] Sent: Wednesday, March 07, 2012 4:24 PM To: [email protected] Subject: RE: [cas-user] Stress testing CAS in production Thanks for the feedback Marv. We were using jmeter against our test CAS (1 or 2 nodes) where were able to get about 800 logins in 50 secs before things (getting the CAS login to render) got slow. I will look into the jmeter/cas link you sent. I think our biggest issue was that this application opened at noon. So we went from about 25 sessions to well over 2000 in the first 5 seconds. ________________________________________ From: Marvin Addison [[email protected]] Sent: Wednesday, March 07, 2012 3:03 PM To: [email protected] Subject: Re: [cas-user] Stress testing CAS in production > How many logins per hour can CAS handle? As many as your infrastructure supports ;) Seriously, it may be helpful to distinguish between logins and service accesses since login is far more resource intensive generally than simply granting and validating a ticket for service access (assuming no proxies, and even then still probably lighter weight). > So we just did 14k logins to access this app in 1 hr. CAS did “slow”, login > page slow to appear and service ticket expirations. We can easily handle 10 auths/sec per node, which scales to 36000/hr per node, on our JPA backend. You're well below where 4 nodes should be straining in my opinion, but that assumes the other system components (LDAP, Oracle) can scale linearly with your Java app servers. That is a fairly dubious assumption in my experience. I encourage you to do some testing yourself using one or more of the JMeter tests attached to https://wiki.jasig.org/display/CASUM/Apache+JMeter. If I had to guess, I'd imagine your database is the bottleneck, but some stress testing will help provide some evidence. M -- You are currently subscribed to [email protected] as: [email protected] To unsubscribe, change settings or access archives, see http://www.ja-sig.org/wiki/display/JSG/cas-user -- You are currently subscribed to [email protected] as: [email protected] To unsubscribe, change settings or access archives, see http://www.ja-sig.org/wiki/display/JSG/cas-user -- You are currently subscribed to [email protected] as: [email protected] To unsubscribe, change settings or access archives, see http://www.ja-sig.org/wiki/display/JSG/cas-user
