Long-range dependence and self-similarity in World Wide Web proxy cache references

The existence of long-range dependence and self-similarity properties within World Wide Web proxy cache references is reported. Fractional ARIMA processes are used to generate long-range dependent cache hit/miss sequences in order to model the growth in cache occupancy. This model was compared to a model using independent cache hit/miss sequences and it was found that the results predicted by these models differed significantly. However, the results of the fractional ARIMA model resembled the growth of an actual cache. Trace simulations were carried out in order to investigate the effects of long-range dependence and self-similarity on cache replacement and garbage collection policies