In the bottom chart of Figure 3 we can see that the view provided by the ten second data is very
misleading. In that view it appears that the variation in response time
is only from 1.25s per request to 2.75s. However, in the top chart which was
sampled at once per second it becomes clear that the slow requests are as long as
six or seven seconds in some cases and typically at least four seconds. This is
because the requests that were pending when the garbage collection starts are all
biased by the full length of the garbage collection interval. Slow sampling hides
many such behaviors as this, resulting in deployments that don't perform as expected.
Using the best most accurate tools makes for more informed and accurate conclusions
about the readiness of an application for production. WHAM provides the accuracy
that organizations need to deploy well-tuned applications. Our Resolver tool is
just one of the tools we can offer to your development and test teams.
Figure 3
