Sun Microsystems, Inc.

Hariprasad B Mankude

Sohrab Modi

Sankar Ramamoorthi

Mahalingam Mani

Kevin C Fox

Brian M Oki

One embodiment of the present invention provides a system that facilitates forwarding fragments of a packet received from a source node to a destination node, wherein the destination node is determined based upon information within a first fragment of the packet. The system operates by receiving at least one fragment of the packet at an interface node from the source node. The system uses a packet identifier from the fragment to look up an entry for the packet within a packet forwarding data structure. If this entry specifies the destination node, the system forwards the fragment to the destination node. If the entry does not specify the destination node, and if the fragment is not the first fragment of the packet, the system links the fragment into the entry for the packet within the packet forwarding data structure, so that the fragment can be forwarded to the destination node when the destination node later becomes known. In one embodiment of the present invention, if the entry for the packet does not specify the destination node, and if the fragment is the first fragment of the packet, the system: determines the destination node from information contained within the first fragment; initializes the entry for the packet to specify the destination node; and forwards any fragments linked into the entry as well as the first fragment to the destination node.

Method and apparatus for forwarding packet fragments

A scalable cluster system that provides scalable services for client applications is provided with a forwarding list. The scalable services are transparent to the client application. To facilitate this transparent scalability, the system provides a forwarding list. For various operational reasons, such as tuning the system, an operator may change the work distribution weights between nodes of a scalable cluster system. Such a change in work distribution weights may change how packets are assigned to nodes. Forwarding lists are provided so that if the work distribution weights are changed while there are existing connections the forwarding lists allow packets from existing connections to go to the same node as earlier packets from the same connection.

Method for creating forwarding lists for cluster networking

One embodiment of the present invention provides a system for forwarding a packet between nodes in a clustered computing system. The system operates by receiving the packet at an interface node in the clustered computing system. This packet includes a source address specifying a location of a client that the packet originated from, and a destination address specifying a service provided by the clustered computing system. The system selects a server node in the clustered computing system to send the packet to from a plurality of server nodes that are able to provide the service. Next, the system forwards the packet to the server node so that the server node can provide the service to the client by, attaching a transport header to the packet, the transport header containing an address of the server node, and sending the packet to the server node through an interface. This interface is used for communications between the interface node and other nodes in the clustered computing system. In one embodiment of the present invention, in forwarding the packet to the server node, the system load balances between multiple redundant paths between the interface node and the server node.

Method and apparatus for fast packet forwarding in cluster networking

One embodiment of the present invention provides a system that uses a packet distribution table to distribute packets to server nodes in a cluster of nodes that operate in concert to provide at least one service. The system operates by receiving a packet at an interface node in the cluster of nodes. This packet includes a source address specifying a location of a client that the packet originated from, and a destination address specifying a service provided by the cluster of nodes (and possibly a protocol). The system uses the destination address to lookup a packet distribution table. The system then performs a function that maps the source address to an entry in the packet distribution table, and retrieves an identifier specifying a server node from the entry in the packet distribution table. Next, the system forwards the packet to the server node specified by the identifier so that the server node can perform a service for the client. In this way, packets directed to a service specified by a single destination address are distributed across multiple server nodes in a manner specified by the packet distribution table. In one embodiment of the present invention, the system allows the server node to send return communications directly back to the client without forwarding the communications through the interface node. In one embodiment of the present invention, the function includes a hash function that maps different source addresses to different entries in the packet distribution table in a substantially random manner. Note that this hash function always maps a given source address to the same entry in the packet distribution table.

Method and apparatus for providing scalable services using a packet distribution table

One embodiment of the present invention provides a system that allows an I/O request to proceed when a primary server that is processing the I/O request fails, and a secondary server takes over for the primary server. Upon receiving an I/O request from an application running on a client, the system stores parameters for the I/O request on the client, and sends the I/O request to the primary server. Next, the system allows the application on the client to continue executing while the I/O request is being processed. If the primary server fails after the I/O request is sent to the primary server, but before an I/O request completion indicator returns from the primary server, the system retries the I/O request to the secondary server using the parameters stored on the client. The I/O request may originate from a number of different sources, including a file system access, an I/O request from a database system, and a paging request from a virtual memory system. In a variation on the above embodiment, the act of storing the parameters for the I/O request on the client includes creating a distributed object defined within a distributed object-oriented programming environment, and sending a reference to the distributed object to the primary server to be stored on the primary server. This causes a distributed operating system to keep track of the reference so that if the primary server fails, the reference count on the distributed object drops to zero and the distributed operating system notifies the client that the distributed object is unreferenced. This allows the client to deduce that the primary server has failed.

Growing community of inventors

Fremont, CA, United States of America

Hariprasad B Mankude