#### DMCA

## Service Differentiation for Delay-Sensitive Applications: An Optimisation-Based Approach (2002)

### Cached

### Download Links

- [research.microsoft.com]
- [www.thlab.net]
- [research.microsoft.com]
- [research.microsoft.com]
- [www.thlab.net]
- [research.microsoft.com]
- DBLP

### Other Repositories/Bibliography

Venue: | Tech. Rep., Microsoft Research |

Citations: | 2 - 1 self |

### Citations

1943 |
Rate control for communication networks: Shadow prices, proportional fairness and stability
- Kelly, Maulloo, et al.
- 1998
(Show Context)
Citation Context ...or Delay-Sensitive Users We now describe a framework which identifies user preferences with multidimensional utility functions. We make use of the resource pricing framework described by Kelly et al. =-=[1,13]-=- which requires the network to send feedback signals (or charges) to users, perhaps conveyed by a packet marking strategy, that reflect the 4 Is is possible, however, that EDF would provide a signific... |

409 |
Network Calculus: A theory of deterministic queues for the Internet. Number 2050 in LNCS.
- Boudec, Thiran
- 2002
(Show Context)
Citation Context ...eadline first scheduler, the required capacity is cEDF ¥ max ¢ ρ1 ¦ ρ2 £ σ1§ d1£ 3 ¢ σ1 ¦ σ2 ¦ ¢ d2 � d1¤ ρ1¤¨§ d2¤©�sThese results follow by applying a result of [7] (see al=-=so Theorem 2.3.2, p.99 of [8]). When d2-=- is large and σ1§ d1 is larger than ρ1 ¦ ρ2, we find that the ratio of these quantities is 1 ¦ σ2§ σ1, which becomes significant if the burstiness parameter σ2 is not negligible compared to ... |

350 | Resource pricing and the evolution of congestion control. Automatica 35
- Gibbens, Kelly
- 1998
(Show Context)
Citation Context ...lgorithms with simple congestion control algorithms is studied via simulations. 1 Introduction With a single best-effort service class it is possible to offer a simple form of service differentiation =-=[1]-=-, whereby users who pay a higher price achieve a proportionally higher transmission rate. However, the applicability of such techniques is limited 1 Partially supported by the National Science Foundat... |

295 |
Performance Guarantees in Communication Networks.
- Chang
- 2000
(Show Context)
Citation Context ...e deadlines. The comparison of cFIFO and cEDF is thus indicative of the maximal gain that can be achieved by introducing multiple classes in the network. Using the results of Example 2.3.13, p. 61 in =-=[9], one also sees that f-=-or priority scheduling, the minimal capacity needed is: ¥ cPriority max ¢ ¢ ρ2£ σ1§ d1£ ρ1 ¦ ρ1 ¦ σ2¤¨§ d2¤©� σ1 ¦ We thus see that when d2 is large, this coincides with cEDF, wh... |

206 |
Elements of Queueing Theory,
- Baccelli, Br6maud
- 1994
(Show Context)
Citation Context ...£ ¢ are delayed by an amount σ, while w¤ � the workload at time T ¢ is exactly σ hence the impact on later arriving w¤ packets does not depend on w. Applying the integration formula (8.3.3), =-=p. 49 in [18] ¢ yields the �-=-�� ρ¤ expression 1 for the middle-term in the right-hand side of this ¢ wρ§ ¢ expression. The constant K is then determined by using the expression ¦ σ¤¨¤ ∂x xE W for the 13saverage s.p.s... |

175 | Exact admission control for networks with a bounded delay service,”
- Liebeherr, Wrege, et al.
- 1996
(Show Context)
Citation Context ... on the relative delay requirements and the burstiness of the taffic. We now review these results for a simple leaky-bucket traffic profile. Suppose that arrivals of type i are ¢ σi£ ρi¤ -constra=-=ined [7]. The capacity c-=- needed to handle these arrivals without loss and within the deadlines is, in the FIFO case, ¥ cFIFO max ¢ ¢ ρ2£ ¦ ¦ σ2¤¨§ d1¤©£ ρ1 σ1 while with an earliest deadline first scheduler, ... |

121 |
Open Queueing Networks in Heavy Traffic
- Reiman
- 1984
(Show Context)
Citation Context ... we need to ensure that P ¢ W ¦ σ ¡ d1¤�� θ. In heavy traffic, the distribution of W is close to that of 4sλσ2 ¢ ¢ � § ρ¤�¤ 2 1 X, where X is an exponential random variable with =-=unit mean 3 (see e.g. [11]), thus P ¢ X ¡ � x¤�� θ x ¡ ��-=-� logθ. Hence, we must ensure ρ1 ¦ ρ2 � 2 ¢ � σ¤ � d1 σlog ¢ θ¤�¦ 2 ¢ � σ¤�� d1 We now consider preemptive priority scheduling. Class 1 packets do not see class 2 packets.... |

120 | Deterministic delay bounds for VBR video in packet-switching networks: Fundamental limits and practical tradeoffs - Wrege, Knightly, et al. - 1996 |

55 | ABE: providing a low-delay service within best effort.
- Hurley, Boudec, et al.
- 2001
(Show Context)
Citation Context ...either class with the elastic-user strategy. 20seach class and whether members of the low priority class are treated fairly. We make the assumption, similar to Alvarez and Hajek [6] and Hurley et al. =-=[19]-=-, that users of the low priority class are delay tolerant, but seek to maximize throughput whereas high priority users seek low delay, perhaps at the expense of lower throughput. Our criteria for ince... |

46 | Real-time queues in heavy traffic with earliest-deadline-first queue discipline
- Doytchinov, Lehoczky, et al.
- 2001
(Show Context)
Citation Context ...r small θ. 5 (1) (2)sor equivalently ρ2 � 2 ¢ � ρ1¤ � 1 σlog ¢ θ¤�¦ 2 ¢ � ρ1¤ 1 ¢ � σ¤ d2 ¢ � σ¤ d2 � ρ1� (3) The analysis of Earliest Deadline First scheduling i=-=n heavy traffic has been presented in [12]-=-. We follow the approach of [12] and consider preemptive resume EDF scheduling, where all customers are accepted into the system, and can thus have negative deadlines. Applying the results of Section ... |

32 | Fundamental Trade-offs in Aggregate Packet Scheduling - Zhang, Duan, et al. |

26 | On packet marking at priority queues.
- Gibbens, Kelly
- 2002
(Show Context)
Citation Context ...sending jointly at rates xr1 and xr2 in classes 1 and 2 takes the form ¢ xr1£ xr2£ D1£ D2¤�¥ Ur Vr 3.2 Loss–based congestion cost ¢ xr1£ xr2¤�� xr1D1 � xr2D2. ¢ Alternatively, one =-=can consider (as in [14,1]) the r-=-ate of packet x� x¤ loss events, ¢ , as the cost of � x¤ congestion, where is the packet loss probability. This might be more appropriate in the situation where user utilities do not depend on ... |

15 | Is Service Priority Useful in Networks
- Bajaj, Breslau, et al.
- 1998
(Show Context)
Citation Context ...es to perform rate control necessary to satisfy their collective delay requirements. The question of whether delay-sensitive applications require service differentiation was addressed by Bajaj, et al =-=[5]-=-, who found the answer depends on the adaptive behaviour of the applications themselves and also on the burstiness of traffic. We study rate-adaptive control rather than the delay-adaptive behaviour c... |

8 | Traffic shaping for end-to-end delay guarantees with EDF scheduling - SIVARAMAN, CHIUSSI, et al. |

5 | On using marks for pricing in multiclass packet networks to provide multidimensional QoS,” Submitted
- Alvarez, Hajek
- 2001
(Show Context)
Citation Context ...n the burstiness of traffic. We study rate-adaptive control rather than the delay-adaptive behaviour considered in [5], but reach similar conclusions about the effect of burstiness. Alvarez and Hajek =-=[6]-=- also consider the necessity of multiple classes in the case of a mixed population of rate-sensitive and loss-sensitive users within the pricing framework of Gibbens and Kelly [1], which we also adopt... |

3 | How to mark fairly, in: Workshop on Internet Service Quality - Wischik - 1999 |

1 |
project COST 224, Performance evaluation and design of multiservice networks, European Community
- unknown authors
- 1992
(Show Context)
Citation Context ...Using this approximation, we therefore need to ensure that 2 ¢ 1 � ρ¤ ¢ ¢ ρ1¤ � � σ¤ 1 λσ d2 2 logθ£ � � 3 The exact distribution of the unfinished work in an M/D/1 queue is giv=-=en in explicit form in [10], p.112. -=-Using heavy traffic approximation rather than the exact distribution affects only marginally the schedulable regions we derive, at least for small θ. 5 (1) (2)sor equivalently ρ2 � 2 ¢ � ρ1¤ ... |

1 |
Discrete Event Systems, Analysis and Control
- Vazquez-Abad
- 2000
(Show Context)
Citation Context ...eriod, the sample path shadow price for 6 A rigorous proof of this fact typically involves a dominated convergence argument, and dominating variables have to be found for each specific situation; see =-=[16] for more details. -=-11 £ (8)spacket m is exactly N � ∑ n� 1 Zn � Z � � m� � n � Applying the cycle formula, it is then seen that the corresponding average packet price for packets of a given type, say ... |

1 |
Analysis of a single server atm queue with priority scheduling
- Walraevens, Bruneel
- 1999
(Show Context)
Citation Context ...t busy period when it enters service. The amount by which this overestimates the correct packet price on average can be computed exactly in the case of an M/D/1 queue, using, for instance, results of =-=[17]-=-). We find that the resulting average price is never more than twice the correct price, and approaches the correct price as load increases to 1. An alternative approach is to charge each packet accord... |