Mechanism for implementing class redirection in a cluster

• US 20030103455A1
• Filed: 11/30/2001
• Published: 06/05/2003
• Est. Priority Date: 11/30/2001
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Pinto, Oscar P.
• Granted Patent: US 7,099,337 B2
• Time in Patent Office: Days
• US Class Current: 370/230
• CPC Class Codes: H04L 45/02 Topology update or discovery
• CPC Class Codes: H04L 47/10 Flow control; Congestion control
• CPC Class Codes: H04L 49/10 characterised by the switching fabric construction

Abstract

A mechanism is provided at a host node to implement redirection for Class Managers that do not reside on the host node in order to process incoming data messages accordingly in a switched fabric for scalable solutions.

Claims

1. A method for implementing class redirection on a host node in a switched fabric, comprising:

storing a registration message including a Local Identifier (LID) of a Class Manager for a given class of general services from a remote node across the switched fabric, upon registration by the Class Manager;

determining if redirection information for a registered class is included in the registration message, upon receipt of an incoming data message from the switched fabric; and

redirecting the incoming data message to a stored Local Identifier (LID) of the Class Manager for the registered class at the remote node across the switched fabric, if redirection information for the registered class is included in the registration message.

2. The method as claimed in claim 1, wherein the host node serves as a subnet management (SM) node providing management services, including basic initialization such as discovering fabric topology, assigning Local Identifiers (LID) to all ports that are connected to the switched fabric, programming switch forwarding tables and maintaining general functioning of the switched fabric to provide access to information such as data paths and alternate data paths between end nodes, topology change notifications and notification of events, including error detection, and recovery procedures.

3. The method as claimed in claim 1, wherein the redirection information is part of a Management Datagram (MAD) including a ClassPortInfo attribute used to indicate if redirection is supported at the host node and contain all information necessary to implement class redirection on the host node.

4. The method as claimed in claim 1, wherein the incoming data message is dropped if redirection information for the registered class is not included in the registration message, and the host node has no Class Manager resident thereon to process the incoming data message.

5. The method as claimed In claim 1, wherein the host node serves as a subnet management (SM) node including one or more Class Managers such as Subnet Administrator (SA), Name Services Manager, Baseboard Manager, 10 Resource Manager (JORM), Device Manager, Vendor Specific Manager and Application Specific Manager for sending and receiving date messages on a special queue pair (QP1) known as General Service Interface (GSI) in accordance with the “

InfiniBand™

Architecture Specification”

.

6. The method as claimed in claim 1, wherein the host node is a switch which stores the registration message including the Local Identifier (LID) of the Class Manager for a given class of general services downloaded from the remote node across the switched fabric, in Class redirection tables, and redirects the incoming data message to the stored Local Identifier (LID) of the Class Manager for the registered class at the remote node across the switched fabric from the Class redirection tables, if redirection information for the registered class is included in the registration message.

7. The method as claimed in claim 6, wherein the Class redirection tables are implemented to hold redirection LID of a given class and to support a get and set operation on the Class redirection tables.

8. A host node, comprising:

at least one channel adapter (CA) including one or more ports to support data transfers, via a subnet; and

an access module including a General Services Agent (GSA) to enable one or more entities to send and receive data messages of management services on the host node, via the subnet, and a Class Redirection Manager to implement class redirection for Class Managers that do not reside on the host node.

9. The host node as claimed in claim 8, wherein the Class Redirection Manager is configured to:

store a registration message including a Local Identifier (LID) of a Class Manager for a given class of general services from a remote node, via the subnet, upon registration by the Class Manager;

determine if redirection information for a registered class is included in the registration message, upon receipt of an incoming data message from the subnet; and

redirecting the incoming data message to a stored Local Identifier (LID) of the Class Manager for the registered class at the remote node, via the subnet, if redirection information for the registered class is included in the registration message.

10. The host node as claimed in claim 9, wherein the General Service Agent (GSA) is configured to provide management services, including basic initialization such as discovering fabric topology, assigning Local Identifiers (LID) to all ports that are connected to the subnet, programming switch forwarding tables and maintaining general functioning of the subnet to provide access to information such as data paths and alternate data paths between end nodes, topology change notifications and notification of events, including error detection, and recovery procedures.

11. The host node as claimed in claim 9, wherein the redirection information is part of a Management Datagram (MAD) including a ClassPortInfo attribute used to indicate if redirection is supported at the host node and contain all information necessary to implement class redirection on the host node.

12. The host node as claimed in claim 9, wherein the incoming data message is dropped if redirection information for the registered class is not included in the registration message, and the host node has no Class Manager resident thereon to process the incoming data message.

13. The host node as claimed in claim 9, wherein the access module further comprises one or more Class Managers such as Subnet Administrator (SA), Name Services Manager, Baseboard Manager, IO Resource Manager (IORM), Device Manager, Vendor Specific Manager and Application Specific Manager for sending and receiving date messages on a special queue pair (QP1) known as General Service Interface (GSI) in accordance with the “

InfiniBand™

Architecture Specification”

.

14. The host node as claimed in claim 9, further comprising:

a plurality of switches each of which stores the registration message including the Local Identifier (LID) of the Class Manager for a given class of general services downloaded from the host node, via the subnet, in Class redirection tables, and redirects the incoming data message to the stored Local Identifier (LID) of the Class Manager for the registered class at the host node, vias the subnet, from the Class redirection tables, if redirection information for the registered class is included in the registration message.

15. The host node as claimed in claim 14, wherein the Class redirection tables are implemented to hold redirection LID of a given class and to support a get and set operation on the Class redirection tables.

16. The host node as claimed in claim 110, wherein the General Services Agent (GSA) is provided for management services including a subnet administration service which provides data path information to reach fabric-attached devices;

a communication management service which provides the means to set up and manage communications between queue pairs (QP);

a performance management service which specifies a set of facilities for examining various performance characteristics of the subnet;

a device management service which specifies the means for determining the type and location of various types of subnet-attached devices;

a device configuration service which assigns subnet-attached devices to the host node;

a baseboard management service which allows management of the subnet-attached devices; and

a network protocol service which specifies mechanisms to support transport of Simple Network Management Protocol “

SNMP”

operations through the subnet.

17. A computer readable medium comprising instructions that, when executed by a host node in a switched fabric including end nodes and switches interconnected via links, cause the host node to implement class redirection by performing the steps of:

storing a registration message including a Local Identifier (LID) of a Class Manager for a given class of general services from a remote node across the switched fabric, upon registration by the Class Manager;

determining if redirection information for a registered class is included in the registration message, upon receipt of an incoming data message from the switched fabric; and

redirecting the incoming data message to a stored Local Identifier (LID) of the Class Manager for the registered class at the remote node across the switched fabric, if redirection information for the registered class is included in the registration message.

18. The computer readable medium as claimed in claim 17, wherein the host node is a switch configured to store the registration message including the Local Identifier (LID) of the Class Manager for a given class of general services downloaded from the remote node across the switched fabric, in Class redirection tables, and redirect the incoming data message to the stored Local Identifier (LID) of the Class Manager for the registered class at the remote node across the switched fabric from the Class redirection tables, if redirection information for the registered class is included in the registration message.

19. The computer readable medium as claimed in claim 18, wherein the redirection information is part of a Management Datagram (MAD) including a ClassPortInfo attribute used to indicate if redirection is supported at the host node and contain all information necessary to implement class redirection on the host node.

20. The computer readable medium as claimed in claim 18, wherein the incoming data message is dropped if redirection information for the registered class is not included in the registration message, and the host node has no Class Manager resident thereon to process the incoming data message.