Buffer Overflows of Merging Streams

We consider a network merging streams of packets with different quality of service (QoS) levels, where packets are transported from input links to output links via multiple merge stages. Each merge node is equipped with a finite buffer, and since the bandwidth of a link outgoing from a merge node is in general smaller than the sum of incoming bandwidths, overflows may occur. QoS is modeled by assigning a positive value to each packet, and the goal of the system is to maximize the total value of packets transmitted on the output links. We assume that each buffer runs an independent local scheduling policy, and analyze FIFO policies that must deliver packets in the order they were received. We show that a simple local on-line algorithm called Greedy does essentially as well as the combination of locally optimal (off-line) schedules. We introduce a concept we call the weakness of a link, defined as the ratio between the longest time a packet spends in the system before transmitted over the link, and the longest time a packet spends in that link’s buffer. We prove that for any tree, the competitive factor of Greedy is at most the maximal link weakness.

[1]  Robert E. Tarjan,et al.  Amortized efficiency of list update and paging rules , 1985, CACM.

[2]  Van Jacobson,et al.  Random early detection gateways for congestion avoidance , 1993, TNET.

[3]  Boaz Patt-Shamir,et al.  Greedy Packet Scheduling on Shortest Paths , 1993, J. Algorithms.

[4]  Walter Willinger,et al.  On the self-similar nature of Ethernet traffic , 1993, SIGCOMM '93.

[5]  Srinivasan Keshav,et al.  An Engineering Approach to Computer Networking: ATM Networks , 1996 .

[6]  David Clark,et al.  An Approach to Service Allocation in the Internet , 1997 .

[7]  David L. Black,et al.  An Architecture for Differentiated Service , 1998 .

[8]  Miguel A. Labrador,et al.  Packet dropping policies for ATM and IP networks , 1999, IEEE Communications Surveys & Tutorials.

[9]  Christophe Diot,et al.  Simple performance models of differentiated services schemes for the Internet , 1999, IEEE INFOCOM '99. Conference on Computer Communications. Proceedings. Eighteenth Annual Joint Conference of the IEEE Computer and Communications Societies. The Future is Now (Cat. No.99CH36320).

[10]  Andras Veres,et al.  The chaotic nature of TCP congestion control , 2000, Proceedings IEEE INFOCOM 2000. Conference on Computer Communications. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat. No.00CH37064).

[11]  Yishay Mansour,et al.  Competitive queue policies for differentiated services , 2000, Proceedings IEEE INFOCOM 2000. Conference on Computer Communications. Nineteenth Annual Joint Conference of the IEEE Computer and Communications Societies (Cat. No.00CH37064).

[12]  Allan Borodin,et al.  Adversarial queuing theory , 2001, JACM.

[13]  Boaz Patt-Shamir,et al.  Buffer overflow management in QoS switches , 2001, STOC '01.

[14]  Yishay Mansour,et al.  Competitve buffer management for shared-memory switches , 2001, SPAA '01.

[15]  Yishay Mansour,et al.  Loss-bounded analysis for differentiated services , 2001, ACM-SIAM Symposium on Discrete Algorithms.

[16]  Yishay Mansour,et al.  Harmonic buffer management policy for shared memory switches , 2002, Proceedings.Twenty-First Annual Joint Conference of the IEEE Computer and Communications Societies.

[17]  Boaz Patt-Shamir,et al.  Nearly optimal FIFO buffer management for DiffServ , 2002, PODC '02.

[18]  Management of multi-queue switches in QoS networks , 2003, STOC '03.

[19]  Rafail Ostrovsky,et al.  Dynamic routing on networks with fixed-size buffers , 2003, SODA '03.