Browse Prior Art Database

Method and System for Optimizing Multicast Management in Infiniband (IB) Networks Disclosure Number: IPCOM000239819D
Publication Date: 2014-Dec-03
Document File: 2 page(s) / 38K

Publishing Venue

The Prior Art Database


A method and system is disclosed for optimizing multicast management in infiniband (IB) networks.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 01 of 2

Method and System for Optimizing Multicast Management in Infiniband (IB) Networks

Infiniband (IB) is a computer network communication link used in high-performance computing, that enables seamless data interconnection between and within computers connected in network. IB standard requires presence of software entities on at least one of the nodes connected in IB network. Software entities such as, for example, Subnet Manager (SM) performs task of assigning addresses and routing table configurations. Applications that use existing multicast

management face several problems that arise from election of a new SM when a previous instance fails. This is a valid scenario that may happen because a node running SM instance is rebooted or has a hardware or software failure. Since the definition of existing multicast groups and knowledge of which ports are joined exists only in the memory of SM

process, the new SM resets all switch multicast forwarding tables resulting in severe delays. So a method is needed for optimizing multicast management in IB networks that can avoid data traffic interruption and would consider restart or

reelect of SM as a least effective event.

Disclosed are two methods and systems for optimizing multicast management in infiniband (IB) networks. The first method and system utilizes one or more multicast groups with a predefined multicast local address (MLID) assigned to each of the one or more multicast groups. These static multicast group definitions include other required group parameters like QKEY, MGID, etc. Software entities (ex: Subnet Manager (SM)) and software applications that use multicast management share predefined multicast group definitions. Static multicast definitions are used for enabling communications between software applications that are expected to run on each endpoint (example of important applications include IP over IB and Ethernet over IB protocols). Thereafter, each software application configures a QP and an address handle for joining them to a static definition of QKEY, multicast global identifier (MGID) and MLID but is not required to send JOIN requests to SM . Thereafter, the SM joins relevant switch ports to the predefined multicast groups as described below.

In accordance with the first method and system, switch ports can toggle between states such as active and inactive. The SM process considers all active end points as joined, an inactive end points as un-joined for multicast routing

configuration and reference counting. When a switch port is active and connects to an end point, the SM joins the switch port to the predefined MLIDs. However when a switch port is not active, the SM removes the switch port from the predefined MLIDs. Further,...