aboutsummaryrefslogtreecommitdiffstats
path: root/resources/libraries/python/PLRsearch/PLRsearch.py
diff options
context:
space:
mode:
Diffstat (limited to 'resources/libraries/python/PLRsearch/PLRsearch.py')
-rw-r--r--resources/libraries/python/PLRsearch/PLRsearch.py14
1 files changed, 11 insertions, 3 deletions
diff --git a/resources/libraries/python/PLRsearch/PLRsearch.py b/resources/libraries/python/PLRsearch/PLRsearch.py
index cdfd308149..ce65fd2ec8 100644
--- a/resources/libraries/python/PLRsearch/PLRsearch.py
+++ b/resources/libraries/python/PLRsearch/PLRsearch.py
@@ -426,13 +426,21 @@ class PLRsearch:
Integrator assumes uniform distribution, but over different parameters.
Weight and likelihood are used interchangeably here anyway.
- Each trial has an offered load, a duration and a loss count.
- Fitting function is used to compute the average loss per second.
- Poisson distribution (with average loss per trial) is used
+ Each trial has an intended load, a sent count and a loss count
+ (probably counting unsent packets as loss, as they signal
+ the load is too high for the traffic generator).
+ The fitting function is used to compute the average loss rate.
+ Geometric distribution (with average loss per trial) is used
to get likelihood of one trial result, the overal likelihood
is a product of all trial likelihoods.
As likelihoods can be extremely small, logarithms are tracked instead.
+ The current implementation does not use direct loss rate
+ from the fitting function, as the input and output units may not match
+ (e.g. intended load in TCP transactions, loss in packets).
+ Instead, the expected average loss is scaled according to the number
+ of packets actually sent.
+
TODO: Copy ReceiveRateMeasurement from MLRsearch.
:param trace: A multiprocessing-friendly logging function (closure).