Poisson distributed random traffic for fixed average lambda would be more realistic. If you really want, you can keep a rolling sum of your poisson samples thus

Poisson distributed random traffic for fixed average lambda would be more realistic. If you really want, you can keep a rolling sum of your poisson samples thus far (which translates into current simulated time), then use a sin function to adjust lambda before taking the next sample. This simulates day/night traffic or any other predictable "seasonal" effect.