| This thesis explores the self-similarity in Web traffic and its relationship to workload parameters, including request size, file size, embedded references, off times, and file composition. SURGE (Scalable URL Reference Generator) was used to execute approximately 600 simulations where the workload parameters were varied between 20% and −30%. We examined the effects on the objects associated with the workload parameters, and the influence by each on self-similarity in the simulated traffic, as represented by the Hurst value. The relationship between the workload and self-similarity was described by a sensitivity index (the ratio between the percentage change in the Hurst value and the parameter object). The file size parameter resulted in the highest sensitivity index of −5.l. The request size was the dominant parameter since it had the greatest, single parameter impact on the degree of self-similarity. Request size was also the object of secondary effects from other workload parameter manipulations. The off time workload parameter produced the highest percentage change, but the self-similarity impact was small.; The simulation tool was crucial in achieving our goal to quantify self-similarity sensitivity. We searched for a reliable tool that provides for parametric manipulations. SURGE offered the necessary flexibility, but reliability had not been validated. Therefore, we spent considerable effort verifying that it would meet our needs. We assessed its repeatability, precision, ability to reproduce real world scenarios, and resource efficiency. We discovered that it was reliable in all aspects, and, we noted the importance of using it properly in order to achieve the desired workloads, which were based on our studies of real-world systems. |
