J.-Y. Le Boudec and A. Charny. Packet scale rate guarantee for non-FIFO nodes. In Proceedings of INFOCOM 2002.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....a distributed admission control scheme with no per flow state. We could have provided a stronger service that achieves worst case per flow delay guarantees but we chose not to do so for the following reasons. First, worst case delays show up only in constructed and unrealistic environments [16, 17]. Common delays in our approach are very close to the round trip propagation delay (as shown in x6.4.6) Second, providing delay bounds increases the computational complexity of packet scheduling from O(1) to O(log n) where n is the number of flows in IntServ and the number of queued packets in ....

J. L. Boudec and A. Charny. Packet scale rate guarantee for non-fifo nodes.

No context found.

Le Boudec, J.-Y., Charny, A. "Packet Scale Rate Guarantee for non-FIFO Nodes", Infocom 2002.

No context found.

J.-Y. Le Boudec and A. Charny, "Packet scale rate guarantee for non-FIFO nodes," Tech. Rep. DSC2001.

....EF network, packet flows are shaped individually at the network ingress points, but are served as aggregates by the nodes in the network. The definition of EF PHB relies on Packet Scale Rate Guarantee (PSRG) 10] defined later) Some sample path properties of PSRG nodes are established, e.g. in [3, 6, 5]. In our prior work [23, 24] we obtained some probabilistic bounds on the performance that apply to PSRG nodes. It is of network engineering interest to have tight bounds on the performance of PSRG nodes, which would enable one to dimension the network such that some notion of quality of service ....

....(resp. departure) times. Denote with L n the size in bits of the packet that arrives at T n . A node is said to o#er PSRG with rate r and latency e, if for all n , T # n V n e; where V 0 = 0, and else V n = max T n , min[V n 1 , T # n 1 ] n 0. An equivalent definition [6] is that for all j e T # j L j 1 . L n k=j 1 # . 1.1) Another node model is Guaranteed Rate (GR) 12] We say that the node is GR with rate r and latency e [12] if, similarly, T # n V n e where V 0 = 0 but the recursion for V n is replaced by V n = max [T n , V ....

[Article contains additional citation context not shown here]

Jean-Yves Le Boudec and Anna Charny, Packet scale rate guarantee for non-fifo nodes, Proc. of IEEE Infocom 2002.

....property by one maximum packet size, but does not increase the packet delay. Conversely, but only for a FIFO node, the rate latency service curve implies GR with rate r and latency e. It follows from this equivalence that the delay bounds in Equation (8) hold for a FIFO GR node; it is shown in [34] that it also holds for non FIFO nodes. Specifically, the packet delay for a flow that is # smooth is bounded by t 0 # r t e (11) For GR nodes that are FIFO per flow, the concatenation result obtained with the service curve approach applies. Specifically, the concatenation of I GR nodes ....

....of I GR nodes (that are FIFO per flow) with rates r i and latencies e i is GR with rate r = in i r i and latency e i (I r , where L max is the maximum packet size for the flow. The term (I r is due to packetizers. For GR nodes that are not FIFO per flow, this result is no longer true [34]. The recursion in (10) can be solved easily, using the properties of max plus algebra. We obtain that GR is equivalent to saying that for all n there is some k a k e (12) which is the dual of (5) with [5] C. DiffServ, Aggregate Scheduling and Adaptive Service Curves The ....

[Article contains additional citation context not shown here]

J.-Y. Le Boudec and A. Charny, "Packet scale rate guarantee for non-FIFO nodes," Tech. Rep. DSC2001.

....EF network, packet flows are shaped individually at the network ingress points, but are served as aggregates by the nodes in the network. The definition of EF PHB relies on Packet Scale Rate Guarantee (PSRG) 10] defined later) Some sample path properties of PSRG nodes are established, e.g. in [3, 6, 5]. In our prior work [23, 24] we obtained some probabilistic bounds on the performance that apply to PSRG nodes. It is of network engineering interest to have tight bounds on the performance of PSRG nodes, which would enable one to dimension the network such that some notion of quality of service ....

....Z , T n Vn e; where V0 0, and else Ln Vn = max Tn,min[Vn ,Tn ] n O. Invited Session on Network Calculus, Fifteenth International Symposium on Mathematical Theory of Networks and Systems, University of Notre Dame, Indiana, USA, August, 12 16, 2002 An equivalent definition [6] is that for all j n V V I ] I . 1.1) k=j l Another node model is Guaranteed Rate (GR) 12] We say that the node is GR with rate r and latency e [12] if, similarly, T n Vn e where V0 = 0 but the recursion for Vn is replaced by Vn max [Tn, Vn 1] 7 0. 1.2) An ....

[Article contains additional citation context not shown here]

Jean-Yves Le Boudec and Anna Charny, Packet scale rate guarantee for non-fifo nodes, Proc. of IEEE Infocorn 2002.

....node, one cannot directly deduce a bound on the complementary distribution of the delay from the complementary distribution of the backlog. However, this is possible for a PSRG node. It is shown that the delay from backlog bound holds for PSRG FIFO nodes [3] and also for non FIFO PSRG in [24]. Proposition 1: Bound on Delay) For a PSRG node with rate , it holds 4Q 4 85 8 h 4R4 8 50 ;0 8E8 h[7] Z (16) 4 8 is a delay incurred by an arbitrary packet that arrives at time . Proof: By Theorem 1 in [3] and Theorem III.1 in [24] the delay for a packet arriving ....

....[3] and also for non FIFO PSRG in [24] Proposition 1: Bound on Delay) For a PSRG node with rate , it holds 4Q 4 85 8 h 4R4 8 50 ;0 8E8 h[7] Z (16) 4 8 is a delay incurred by an arbitrary packet that arrives at time . Proof: By Theorem 1 in [3] and Theorem III.1 in [24], the delay for a packet arriving at time is bounded by 476 8C( simply use this majorization to obtain (16) Notice, combining (16) with Corollary 1, and any upper bound on the steady state complementary distribution of the backlog, we obtain an upper bound on the complementary ....

[Article contains additional citation context not shown here]

Jean-Yves Le Boudec and Anna Charny, "Packet scale rate guarantee for non-fifo nodes," in Proc. of IEEE Infocom 2002, New York, USA, June

No context found.

J.-Y. Le Boudec and A. Charny, "Packet scale rate guarantee for non-fifo nodes," Tech. Rep. DSC2001.

....distribution of the delay from the complementary distribution of the backlog. However, this is possible if the node can be abstracted with PSRG, as is the case for the proposed definition of EF. Indeed, a delay from backlog bound for PSRG nodes is given in [6] for FIFO nodes, and it is proven in [21] that it holds without the FIFO assumption. Proposition 1: For a PSRG node with rate and latency 2 , it holds 1 z l # z 2 ( zd 0 (14) where w is a delay incurred by an arbitrary packet that arrives at time . Proof: By Theorem 1 in [6] and Theorem ....

....holds without the FIFO assumption. Proposition 1: For a PSRG node with rate and latency 2 , it holds 1 z l # z 2 ( zd 0 (14) where w is a delay incurred by an arbitrary packet that arrives at time . Proof: By Theorem 1 in [6] and Theorem III.1 in [21], the delay for a packet arriving at time is bounded by l ( G2 ; then simply use this point wise majorization to obtain (14) Thus, combining (14) with Corollary 1 and any upper bound on the steady state complementary distribution to the backlog (e.g. 3) or (4) we obtain an upper ....

[Article contains additional citation context not shown here]

Jean-Yves Le Boudec and Anna Charny, "Packet scale rate guarantee for non-fifo nodes," Tech. Rep. DSC2001.

....node, one cannot directly deduce a bound on the complementary distribution of the delay from the complementary distribution of the backlog. However, this is possible for a PSRG node. It is shown that the delay from backlog bound holds for PSRG FIFO nodes [3] and also for non FIFO PSRG in [24]. Proposition 1: Bound on Delay) For a PSRG node with rate # and latency # , it holds # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # (16) where # # # # is a delay incurred by an arbitrary packet that arrives at time # . Proof: By Theorem 1 in [3] and Theorem III.1 in [24] ....

....in [24] Proposition 1: Bound on Delay) For a PSRG node with rate # and latency # , it holds # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # # (16) where # # # # is a delay incurred by an arbitrary packet that arrives at time # . Proof: By Theorem 1 in [3] and Theorem III.1 in [24], the delay for a packet arriving at time is bounded by ; simply use this majorization to obtain (16) Notice, combining (16) with Corollary 1, and any upper bound on the steady state complementary distribution of the backlog, we obtain an upper bound on the complementary distribution of the ....

[Article contains additional citation context not shown here]

Jean-Yves Le Boudec and Anna Charny, "Packet scale rate guarantee for non-fifo nodes," in Proc. of IEEE Infocom 2002, New York, USA, June

....packet size, but does not increase the packet delay. Conversely, but only for a FIFO node, the rate latency service curve #### # ### # ## # implies GR with rate # and latency #. It follows from this equivalence that the delay bounds in Equation (8) hold for a FIFO GR node; it is shown in [31] that it also holds for non FIFO nodes. Specifically, the packet delay for a flow that is # smooth is bounded by ### ### # #### # # # # # # (11) For GR nodes that are FIFO per flow, the concatenation result obtained with the service curve approach applies. Specifically, the concatenation of ....

....per flow) with rates # # and latencies # # is GR with rate # # ### # # # and latency # # # # # # ### # ## #### # , where # ### is the maximum packet size for the flow. The term ## # ## #### # is due to packetizers. For GR nodes that are not FIFO per flow, this result is no longer true [31]. The recursion in (10) can be solved easily, using the properties of max plus algebra. We obtain that GR is equivalent to saying that for all # there is some # # ## ### #### ## such that # # # # # # # # # ### # # # # # # (12) which is the dual of (5) with ######## # ## # [5] C. ....

[Article contains additional citation context not shown here]

J.-Y. Le Boudec and A. Charny, "Packet scale rate guarantee for non-FIFO nodes," Tech. Rep. DSC

No context found.

J.-Y. Le Boudec and A. Charny. Packet scale rate guarantee for non-FIFO nodes. In Proceedings of INFOCOM 2002.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC