Each data point was tested 10 times; the curves show the average values with standard error bars to show the experimental variation across tests;
these experimental errors decrease for larger message sizes because the tests run for longer and are less subject to random
OS and cache behavior.