Home · Search · Print · Help 

There are two distinct ways to recycle a Polygraph test.

• Continuing an earlier test is possible using Persistent Working Set feature. With this feature, you can, for example, start a test with a cache already full of old URLs and have Polygraph request those URLs as if they were cached during the current test. Persistent working set is useful when you want to continue an earlier test instead of starting from scratch.

• Repeating an earlier test is possible by forcing Polygraph to use exactly the same random number generator seeds. Same seeds allow you to generate nearly identical URL/request sequences. Unlike Persistent Working Set feature above, this kludge mode does not provide Polygraph with information about previously requested URLs so Polygraph will not, for example, expect a cache hit when requesting an "old" URL for the first time in the second test. Using the same seeds is useful when you want to repeat an earlier test from scratch.

This page talks about common principles behind tests/URLs recycling. Please see Continuing Tests and Repeating Tests pages for documentation specific to those two subjects.

1. Do not do it

In most cases, you do not want to force Polygraph to reuse URLs. By default, Polygraph will use a unique set of URLs for every run. Unique URL sets are convenient because they reduce side-effects that one test might have on another. For example, one does not have to flush (empty) the cache between to tests to get configured hit ratio because the second test will not (by default) reuse old URLs.

If you read this page out of curiosity, stop now and come back only if you really need two tests to use the same URL set.

2. What is [not] reproducible

Workload features explicitly controlled by PGL objects can be reproduced with good precision. For example, two tests with the same object size distribution are likely to yield very similar object size distributions. This is default Polygraph behavior, and no special effort is required to enable it. In other words, Polygraph tests are meant to be reproduceable by default. Reproducibility in this context does not imply reusing things like URL strings or request sequences; it implies reusing things like object size distributions or request interarrival distributions.

It is possible to force Polygraph to use the same URL set or request sequence in two tests using one of the two techniques mentioned above.

In general, it is not possible to repeat exact request submission times or exact network packet sizes. Polygraph does not have much control over those things. Kernel scheduler decisions or TCP stack behavior cannot be reproduced in a general environment. Fortunately, such reproduction is usually unnecessary.

Home · Search · Print · Help