doi:10.1016/S0169-7552(98)00226-8 
Copyright © 1999 Elsevier Science B.V. All rights reserved
Performance of ATM available bit rate for bursty TCP sources and interfering traffic1
Duke P. Honga, b, * and Tatsuya Sudaa
a Department of Information and Computer Science, University of California, Irvine, CA 92697, USA
b Science Applications International Corp., 10260 Campus Point Dr., San Diego, CA 92121, USA
Available online 24 March 1999.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
The Available Bit Rate (ABR) service has been shown to enable persistent, greedy data sources to efficiently utilize ATM network resources. However, the throughput for bursty TCP over ABR data traffic in the presence of interfering traffic may not be as good as for persistent ABR sources without interfering traffic. This paper shows that in a comparison of rate based ABR schemes (EPRCA and ERICA) and a credit based ABR scheme (QFC), QFC performance is significantly better than rate based ABR schemes for bursty data traffic with bursty interfering traffic.
Author Keywords: TCP; ATM; ABR; EPRCA; ERICA; QFC; Internet; WebStone
Fig. 1. Single link network simulation model.
Fig. 2. Parking lot network simulation model.
Fig. 3. Transmission characteristics of interfering traffic A and B for parking lot network model.
Fig. 4. Video interfering traffic model multiplexes 5 traces of 50 s samples from “Star Wars” MPEG at 10 s intervals.
Fig. 5. Effects of increasing data load on the 95th percentile response times for SPECWeb-96 file transfers. Single link model, 10 Mbps interfering traffic, all 10 km links.
Fig. 6. Effects of increasing data load on the maximum switch buffer occupancies for SPECWeb-96 file transfers. Single link model, 10 Mbps interfering traffic, all 10 km links.
Fig. 7. Effects of increasing ON state bit rate on the 95th percentile response times for SPECWeb-96 file transfers. Single link model, 20 Mbps data traffic, all 10 km links.
Fig. 8. Maximum source buffer occupancies versus link distance .Single link model, 25 Mbps data traffic, 10 Mbps interfering traffic.
Fig. 9. Maximum switch buffer occupancies versus link distance. Single link model, 25 Mbps data traffic, 10 Mbps interfering traffic.
Fig. 10. C.d.f. of response times for SPECWeb-96 file transfers. Single link network model, MPEG interfering traffic, 0.8 mean link util, all 10 km links.
Fig. 11. C.d.f. of response times for SPECWeb-96 file transfers with interfering ON–OFF traffic streams A and B. Parking lot network model. Link utilization of 0.8.
Fig. 12. TCP over QFC transmission of 100 kB by 8 sources. Single link model with 20 Mbps interfering traffic.
Fig. 13. TCP over ERICA transmission of 100 kB by 8 sources. Single link model with 20 Mbps interfering traffic.
Fig. 14. Effect of ERICA's Target Utilization (TU) and increasing data load on 95th percentile response times for SPECWeb-96 file transfers. Single link model, 10 Mbps interfering traffic, all 10 km links.
Fig. 15. Effect of finite ERICA switch buffers and increasing data load on the 95th percentile response times for SPECWeb-96 file transfers. 1.0 Target Utilization. Single link model, 10 Mbps interfering traffic, all 10 km links.
Table 1. Cell loss and lossy TCP connection rates for ERICA with Target_Util=1.0
Abstract
Purchase PDF (349 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (2)
Purchase PDF (668 K)
