Qlogic Corporation

Intel Corporation

James Kunz

Ian G Colloff

Frank R Dropps

Thomas R Prohofsky

Mark Alan Owen

Todd M Rimmer

Craig M Verba

William J Gustafson

Lloyd Dickman

Ernest G Kohlwey

Edward C Ross

Mark Debbage

Leonard W Haseman

Michael Heinz

Thomas D Lovett

Mark S Birrittella

Albert S Cheng

Jerald K Alston

Eric R Griffith

Oscar J Grijalva

Methods, apparatus, and systems for implementing hierarchical and lossless packet preemption and interleaving to reduce latency jitter in flow-controller packet-based networks. Fabric packets are divided into a plurality of data units, with data units for different fabric packets buffered in separate buffers. Data units are pulled from the buffers and added to a transmit stream in which groups of data units are interleaved. Upon receipt by a receiver, the groups of data units are separated out and buffered in separate buffers under which data units for the same fabric packets are grouped together. In one aspect, each buffer is associated with a respective virtual lane (VL), and the fabric packets are effectively transferred over fabric links using virtual lanes. VLs may have different levels of priority under which data units for fabric packets in higher-priority VLs may preempt fabric packets in lower-priority VLs. By transferring data units rather than entire packets, transmission of a packet can be temporarily paused in favor of a higher-priority packet. Multiple levels of preemption and interleaving in a nested manner are supported.

Hierarchical/lossless packet preemption to reduce latency jitter in flow-controlled packet-based networks

System, method, and apparatus for improving the performance of collective operations in High Performance Computing (HPC). Compute nodes in a networked HPC environment form collective groups to perform collective operations. A spanning tree is formed including the compute nodes and switches and links used to interconnect the compute nodes, wherein the spanning tree is configured such that there is only a single route between any pair of nodes in the tree. The compute nodes implement processes for performing the collective operations, which includes exchanging messages between processes executing on other compute nodes, wherein the messages contain indicia identifying collective operations they belong to. Each switch is configured to implement message forwarding operations for its portion of the spanning tree. Each of the nodes in the spanning tree implements a ratcheted cyclical state machine that is used for synchronizing collective operations, along with status messages that are exchanged between nodes. Transaction IDs are also used to detect out-of-order and lost messages.

System, method and apparatus for improving the performance of collective operations in high performance computing

Method and system for network communication including identifying a first network port to be taken offline. Before taking the first network port offline, processing any pending packet tag for the first network port. The method further includes taking the first network port offline; storing a packet tag destined for the first network port at the second network port, while the first network port is offline; bringing the first network port online; and routing the packet tag stored at the second network port, while the first network port was offline; wherein the packet tag is routed from the second network port to the first network port.

Method and system for taking a network port offline

Method and system for transmitting a multicast message with one or more packets to a plurality of destinations is provided. The system includes an adapter including an entry port to receive the multicast message from a source for transmission to the plurality of destinations; one or more egress ports of the adapter that transmit one or more packets of the multicast message to the plurality of destinations and receives acknowledgement for the one or more packets from the one or more destinations; and a message manager that monitors the delivery status for one or more packets to the plurality of destinations without using a plurality of dedicated individual connections between each of the plurality of destinations and the source.

Method and system for reliable multicast

Method and system for a network switch element is provided. The switch element includes a plurality of megaports, each megaport uniquely identified by a unique megaport address identifier for network addressing. Each megaport includes a plurality of operational ports, each operational port identified by a unique operational port address identifier. The switch element also includes a local crossbar for communication between the plurality of operational ports, and a shared logic module configured to provide common control of the plurality of operational ports within a megaport to allow operational ports to share resource of a single megaport to route network packets there between. The switch element also includes a global crossbar configured to allow communication between the megaports.

Method and system for network switch element

A switch element and a method for routing packets in an IB Multi Level switch and network is provided. The method includes determining if alternate routing is enabled for a packet; determining an alternate route address for the packet, if alternate routing is enabled; and routing the packet using the alternate route address, if the alternate route address is valid. The switch element includes a routing table in a port that determines a base route address; and if alternate routing is enabled for a packet, the port determines an alternate route address for a packet; and routes the packet using the alternate route address.

Method and system for load balancing in InfiniBand switches and networks

Method and system for a switch element is provided. The method includes (a) receiving a portion of a packet at a port of the switch element; (b) generating a tag based on an estimated packet size obtained from the portion of the packet; (c) sending the tag with the estimated packet size to a transmit segment of the port; (d) selecting a request from among a plurality of pending requests for processing the packet associated with the tag; (e) receiving an actual packet size for the packet in step (a); (f) determining if the actual packet size is different from the estimated packet size; and (g) adjusting an arbitration weight used for selecting the request in step (d), if the actual packet size is different from the estimated packet size.

Method and system for tag arbitration in switches

Method and system for network communication between a first port and second port using plurality virtual lanes provided. The method includes: (a) configuring a threshold value for each of the plurality of virtual lanes; wherein the threshold value defines an amount of data that has to be moved from a receive segment of the second port, before a flow control packet is sent by the second port to the first port; (b) setting a timer value for each of the plurality of virtual lanes; wherein a flow control packet is sent by the second port after the timer expires; (c) monitoring the amount of data removed from the receive segment of the second port; and (c) sending a flow control packet if the amount of data exceeds the threshold value or if the timer set in step (b) has expired.

Method and system for transmitting flow control information

Method and system for load balancing in infiniband switches and networks

A storage area network ('SAN') and a system is provided. The SAN includes, a host bus adapter operationally coupled with a credit extender, wherein the credit extender receives frames from a Fibre Channel network and sends the received frames to the HBA based on buffer space available in the HBA. The HBA notifies other Fibre Channel ports of buffer space available in the credit extender. The HBA sends a signal to the credit extender notifying the credit extender of available buffer space in the HBA. The HBA includes a management port for interfacing the HBA with the credit extender.

Method and system for using an in-line credit extender with a host bus adapter

A method and system for credit recovery of lost frames and R_RDYs in an in-line credit extender is provided. The method includes, comparing received frame count and a programmable counter value when BB_SCs are received; loading the difference between the frame count and the programmed counter value into a buffer and to a first counter that counts each frame that is transmitted; and sending BB-SCs to the local device if there is a match between the first counter value and a second programmable counter value. The method also includes, counting received R_RDYs, wherein a first counter counts the received R_RDYs; setting a flag when a BB_SCr is received; and transmitting BB-SCr when the first counter value is zero and the flag is set. Also, R_RDYs received after BB_SCrs are counted by a second counter and transmitted when the second counter value is non-zero.

Growing community of inventors

Plymouth, MN, United States of America

James Kunz