The impacts of self-similar traffic on network performance has received much attention recently, however the effects of control network on self-similar traffic networks has not been fully investigated yet. Therefore, the objective of this paper is to to understand self-similarity on physical grounds in a realistic network environment. This understanding is important when developing efficient and integrated network frameworks within which end-to-end QoS guarantees are fully supported. We first introduce the techniques that we use to generate self-similar network traffic .Then we compare the performance of an Ethernet segment run with heavy-tail traffic and with exponential traffic. Our study results show that self-similar traffic, compared with traditional short-range dependent models, requires longer queues and thus larger buffers in the control network design.