Browse Prior Art Database

Path-Based Hop-By-Hop Buffer Class Window Flow Control

IP.com Disclosure Number: IPCOM000039006D
Original Publication Date: 1987-Apr-01
Included in the Prior Art Database: 2005-Feb-01
Document File: 5 page(s) / 32K

Publishing Venue

IBM

Related People

Chen, MS: AUTHOR [+3]

Abstract

The present publication discusses how packets and buffers should be classified and then proposes a new classification scheme. The new scheme is compared against several well-known schemes. The discussions assume that flow control is performed in the hop-by-hop manner. Namely, every session is broken up into multiple pipes, one for each link. Each of these pipes is flow-controlled separately by a window mechanism. Each link has its own flow-control mechanism separately from other link flow control operation. This invention is used to reduce the following: heavy congestion at intermediate modes, store and forward buffer deadlocks, and unfair resource allocation. In a buffer class (BC) scheme, buffers and packets are divided into classes. A packet can be stored only in a buffer of its own class.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 30% of the total text.

Page 1 of 5

Path-Based Hop-By-Hop Buffer Class Window Flow Control

The present publication discusses how packets and buffers should be classified and then proposes a new classification scheme.

The new scheme is compared against several well-known schemes. The discussions assume that flow control is performed in the hop-by-hop manner. Namely, every session is broken up into multiple pipes, one for each link. Each of these pipes is flow-controlled separately by a window mechanism. Each link has its own flow-control mechanism separately from other link flow control operation. This invention is used to reduce the following: heavy congestion at intermediate modes, store and forward buffer deadlocks, and unfair resource allocation.

In a buffer class (BC) scheme, buffers and packets are divided into classes. A packet can be stored only in a buffer of its own class. Packets of the same class are flow-controlled together, i.e., their transmissions are controlled by a common window. The most straightforward classification is the Session-Based (SB) scheme. In it, packets of the same session are classified into one class and are controlled by a common window. The flow control in TYMNET is an example of SB classification [1]. In order to clearly present and compare more classifications, we note that there are two key considerations in the design of a buffer classification. They are the buffer requirement and the fairness property. In a BC scheme, the receiver needs to pre-allocate a certain amount of buffers for every class of packets. In the example, Jerky Sliding Window, the receiver pre-allocates (2W-1) buffers for each class. The total amount of buffers pre-allocated is then (2W-1)*Nc, where Nc is the number of classes. In SB scheme, for example, Nc is the number of sessions that go through a link. Therefore, from the viewpoint of buffer requirement, it is desirable to have as few classes as possible. In Systems Network Architecture (SNA), there is the concept of Virtual Route (VR) which is a logical pipe between backbone nodes. Sessions that have the same end points and take the same path can be multiplexed on the same VR. Therefore, if packets are classified according to their VRs, then Nc and buffer requirements can be reduced. This is the Virtual Route Based or RB classification. There is also the Hop-Count (HC) classification. In a HC scheme, packets are classified according to their hop-count information. Two versions of HC are as follows. 1. Classification is based on the number of hops that packets have already travelled. 2. Classification is based on the number of hops that packets have yet to travel. The maximum number of classes in an HC scheme is the length in hops of the longest session in the network. Hence, it is easy to see that Nc in HC is significantly smaller than Nc in either SB and RB classifications. Or, HC classification requires many fewer buffers than SB and RB classifications. The second key consideration in buffer classifica...