On Mon, 4 Jun 2007, Sanjai Narain wrote:
An example of where behavioral tests don't seem appropriate is in testing single points of failure. We cannot fail individual components and links to see whether end-to-end service hold up, can we? An analytic approach is needed. However, this is also not an either/or situation. Behavioral and analytic tests are complementary.
I don't see why not; I mean, obviously you don't want to fail things in production, but what are you doing running your tests in production, anyhow? you want to test your failures in the test cluster.
I do agree that it's good to know ahead of time if something can be failed over; just dumb testing is suboptimal, as it won't show the difference between something with no failover by design and something with a broken failover system; but ultimately, the "go no-go" decision is the most important bit, and dumb testing (at least in text/HTTP based environments where I work) can give you that.
_______________________________________________ lssconf-discuss mailing list lssconf-discuss@inf.ed.ac.uk http://lists.inf.ed.ac.uk/mailman/listinfo/lssconf-discuss
