doi:10.1016/j.comnet.2004.12.006
Copyright © 2005 Elsevier B.V. All rights reserved.
A lightweight marker with partial state information for DiffServ networks
Dipartimento di Elettronica, Politecnico di Torino, Torino, Italy
Received 29 July 2004;
revised 29 July 2004;
accepted 6 December 2004.
Responsible Editor: J. Sole-Pareta.
Available online 7 March 2005.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
A problem common to packet markers used in DiffServ architectures is the provision of Quality of Service to TCP flows: specifically, short-lived TCP flows suffer from packet losses at small congestion windows, which inevitably leads to lengthy retransmission timeouts. It would therefore be sensible to mark packets of short-lived flows so as to offer them a greater level of protection. In this work, we propose a class of Fair Markers that, without the penalty of per-flow management, achieves the same performance enhancement of a per-flow marker, but with a much simpler design and limited implementation complexity. Extensive simulations with realistic traffic scenarios and simple analytical models allow us to compare our proposal with existing ones. We also underline the importance of providing a minimum protection to both data and acknowledgment segments of traffic crossing congested domains.
Keywords: DiffServ; TCP; Packet marking; Fairness
Fig. 1. Single-bottleneck network scenario.
Fig. 2. Multi-bottleneck network scenario.
Fig. 3. MRED parameters for edge routers.
Fig. 4. Scheme of the TSW2CM.
Fig. 5. Completion times of HTTP transactions under DS provisioning of servers only: absolute (top) and relative (bottom) values.
Fig. 6. Completion times of HTTP transactions under DS provisioning of servers and clients: absolute (top) and relative (bottom) values.
Fig. 7. Architecture of the fair marker.
Fig. 8. Flow Classifier operations.
Fig. 9. Green (top) and Yellow (bottom) random-marker operations.
Fig. 10. Completion times of HTTP transactions using different markers and considering a TCP-NewReno source with or without ECN: absolute (top) and relative (bottom) values.
Fig. 11. Relative completion times of HTTP transactions using TSW2CM marker (top) and FM3 marker (bottom) when considering TCP-NewReno with ECN versus TCP-NewReno without ECN.
Fig. 12. Completion times of HTTP transactions using different markers: absolute (top) and relative (bottom) values. Multi-bottleneck topology.
Fig. 13. Delay (on the top) and Jitter (on the bottom) of UDP flows with different markers. Multi-bottleneck topology.
Fig. 14. Completion time for different values of the yellow dropping probability.
Fig. 15. Comparison between the proposed model and the simulation measurements for different tft.
Fig. 16. Parameter tuning: SRFT threshold (top) and SRFT size (bottom).
Fig. 17. Number of contemporary active flows measured during simulation and obtained by the model, for different values of the total offered load.
Table 1.
Lengths of client requests

Table 2.
Lengths of server responses
