Multiplex communications – Pathfinding or routing – Switching a message which includes an address header
Reexamination Certificate
1999-04-21
2002-12-31
Marcelo, Melvin (Department: 2663)
Multiplex communications
Pathfinding or routing
Switching a message which includes an address header
C370S233000
Reexamination Certificate
active
06501762
ABSTRACT:
FIELD OF INVENTION
This invention relates generally to intra-class, per-connection schedulers for digital link systems, such as aysnchronous transfer mode (ATM) telecommunications systems, and more particularly, provides a novel scheduler implementing weighted fair queuing by a weight limited first in-first out methodology.
BACKGROUND OF THE INVENTION
Information regarding ATM systems can be found in the ATM Forum, “ATM User-Network Interface Specification,” Version 3.1, September, 1994, and in Martin de Prycker, “Asynchronous Transfer Mode: Solution for Broadband ISDN,” Ellis Horwood, N.Y. 1993, both of which are incorporated herein by reference. Aspects of ATM Systems are disclosed in U.S. Pat. Nos. 5,818,815, 5,794,025, and 5,768,259, also incorporated herein by reference.
An advantage of high bandwidth, fixed length cell based transfer mode techniques, such as ATM, Frame Relay, IP, etc., is the capability of high-speed, low-delay multiplexing and switching of different types of user information traffic—including data, voice, image and video—over the same network, while optimizing bandwidth and resource utilization. In ATM connection-oriented networks, the end-to-end traffic of data for one user, between the source and the destination, is represented by a virtual connection (VC). Several VCs can be grouped into a virtual path (VP) that can be switched as a unit.
An ATM system segments user traffic into fixed-length cells. A header in each cell typically includes a virtual channel identifier (VCI) and a virtual path identifier (VPI) associated with the corresponding cell. The VCI and VPI fields together identify a virtual channel connection that is established when a user requests a network connection in an ATM system. The allocation of available transmission opportunities or slots to user traffic cells is generally referred to as cell scheduling. Devices that accomplish this task are called schedulers.
In a data communication system, such as an ATM system, data can optimally be differentiated by classes (e.g., voice, video, data, and other digital signals). A class scheduler allocates opportunities to the different classes, generally by strict priority or some variation thereof which may incorporate some minimum guarantees to each class. Opportunities, once allocates to a class, must be allocated to connections within the class by a second-order scheduler. A scheduler which operates in this manner is called an intra-class scheduler. An intra-class, per-connection scheduler (i) monitors the status of an input queue for each connection in a given class of service; and (ii) according to some pre-defined criteria, makes decisions about the connection to be serviced at the next transmission opportunity. A digital switching apparatus utilizes these decisions to select a connection for the next transmission opportunity.
The same scheduler can also be used to allocate opportunities of one connection to multiple sub connections according to some pre-defined criteria. Such behavior can be used to schedule virtual channel connections (VCCs) into a virtual path connection (VPC). A scheduler which operates in this manner is called a sub-connection scheduler.
The intra-class per-connection scheduler thus enables the digital link system to determine which connection will utilize the next share of limited bandwidth or processor resources. These decisions are important to the performance of bandwidth-limited digital link systems. Scheduling at the intra-class level of per-connection queues is a major problem to be resolved in cell-based networks.
At the intra-class scheduler level a goal of scheduling is to ensure fairness within a particular class (i.e. attempting to provide transmission opportunities for an equal amount of data for each user while taking into account the relative amount of bandwidth required or subscribed to by a user, that is, that user's “share”). A goal of scheduling is to provide fair access to that share without adding unnecessary delay into the system.
A conventional scheduling algorithm utilized is a Round Robin algorithm. In Round Robin the scheduler polls, or inspects the queue belonging to the connection of each user in a round robin fashion (i.e. one at a time in a predefined repeating order).
There are several disadvantages to the Round Robin algorithm. The scheduler visits the queue belonging to the connection of each user without taking into account the amount of bandwidth to which that user subscribes and/or the amount of data that connection is transmitting. Another disadvantage is that a cell may be required to wait in a per connection queue while the scheduler polls all other per connection queues before that queue is inspected by the scheduler (even if that queue is the only queue attempting to transmit data). Moreover, the length of time a cell must wait is a function of chance. For example, a cell may arrive in a particular per connection queue just as the Round Robin scheduler is about to inspect that queue for available cells, in which case that cell will encounter a relatively low transit delay. However, another cell from the same connection may arrive just after that per connection queue has been inspected (visited), in which case that cell will encounter a relatively high transit delay. Thus, the cell delay variation (CDV) through a Round Robin scheduler is high.
Since CDV is a significant constraint in system design, and since many factors that contribute to CDV are beyond the control of the system design, it is very important that any algorithm designed to ensure fairness does not unnecessarily increase CDV, as the Round Robin algorithm does.
An improvement to the Round Robin system includes the weighted burst Round Robin algorithm. In weighted burst Round Robin the number of contiguous cell opportunities that a connection can utilize when the queue for that connection is polled is dependent upon the relative weight of that connection. Thus, when a connection's queue is polled by the scheduler, the most that connection can transmit is the number of cells equal to its weight (i.e. the connection can send a “burst” of data). Although this algorithm provides a more flexible definition of fairness than the Round Robin, due to its burst nature, the possibility of increased CDV is even greater. In the Round Robin, a maximum of one cell of a user's data is sent when the queue belonging to their connection is visited. However, in the weighted burst Round Robin, a connection of high relative weight can monopolize all cell opportunities for that class for a significant length of time.
Another conventional improvement to the Round Robin algorithm is the weighted interleaved Round Robin algorithm. In this algorithm, connections with a weight greater than 1 are allocated more than one position in the Round Robin pattern. Thus, rather than a single position with multiple cell transmit opportunities, a connection may be provided with multiple smaller bandwidth positions which combined provide the same transmit opportunities spread out over different positions. In this sense, the burst will be spread out over time, somewhat reducing the additional CDV introduced into the system. Unfortunately, some additional CDV is still introduced, which remains a function of chance as to when a cell arrives relative to the inspection of the per connection queue. Further, a weighted interleaved Round Robin algorithm is complex to implement.
Accordingly, a need exists for providing intra-class level scheduling on a weighted fair basis, while limiting the amount of additional CDV introduced to the system.
It is thus an object of the invention to provide relatively low-cost, relatively simple and efficient scheduling mechanisms. It is another object of the invention to provide such mechanisms which minimize the introduction of additional CDV and which provide weighted fairness among connections. It is yet another object of this invention to provide a first in-first out (FIFO) mechanism that preserves the cell arrival order, so that a connection can expect it&apo
Guttman Jacob
Marchand Bernard
Patro Sitaram
Pillar John
Ferris Derrick W
Gibbons Del Deo Dolan Griffinger & Vecchione
Marcelo Melvin
LandOfFree
Scheduler implementing weighted fair queuing by a weight... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Scheduler implementing weighted fair queuing by a weight..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Scheduler implementing weighted fair queuing by a weight... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2995919