Mellanox Technologies, Ltd.

Mellanox Technologies Tlv Ltd.

Nvidia Corporation

Eitan Zahavi

Elad Mentovich

Paraskevas Bakopoulos

Zachy Haramaty

Lion Levi

Noam Bloch

Gil Bloch

Ioannis (Giannis) Patronas

Barak Gafni

Vladimir Zdornov

Liron Mula

Diego Crupnicoff

Benny Koren

Daniel Marcovitch

Dror Goldenberg

Richard Leigh Graham

Amiad Marelli

Alexander Shpiner

Yong Qin

Yaniv Blumenfeld

Aviv Kfir

Idan Burstein

Louis Bennie Capps, Jr

George Elias

Matty Kadosh

Itamar Rabenstein

Sylvie Rockman

Miriam Menes

Yaakov Gridish

Adi Menachem

Dimitrios Syrivelis

Freddy Gabbay

Alex Shpiner

Sagi Kuks

Dror Bohrer

Ori Rottenstreich

Vladimir Koushnir

Eran Aharon

Alex Netes

Aviv Barnea

Prethvi Ramesh Kashinkunti

Julie Irene Marcelle Bernauer

Lior Hodaya Bezen

Gil Levy

Mark Douglas Hummel

Dotan David Levi

Ariel Shahar

Ryan Donovan Wells

Glenn A Dearth

Evelyn Landman

Itshak Kalifa

Roee Moyal

Dimitrios Kalavrouziotis

Michael Lawrence Sabotta

Natan Manevich

Ariel Almog

Yuval Shpigelman

Hillel Chapman

Boris Pismenny

Wojciech Wasko

Ido Bukspan

Alon Webman

Siddha Ganju

Yossef Itigin

Anna Sandomirsky

Hanan Shteingart

Bar Shapira

Tamir Ronen

Ahmad Atamli

Jean-Francois Puget

Daniel Klein

Oded Zemer

Ilya Vershkov

Nikolaos Argyris

Gary Mataev

Tal Anker

Gil Mey-Tal

Michael Tahar

Marina Lipshteyn

Guy Rozenberg Kunievsky

Gal Shohet

Harold Rosenstock

Yuval Eliyahu

Boaz Ben-Nun

Zvi Shmueli

Yuval Leader

Meni Orenbach

Guy Rozenberg

Aleksandr Minchiu

Konstantinos Tokas

Michael Schapira

Nikolaos Terzenidis

Michael Norman

Michael Balint

Zachi Binshtock

Michael Avimelech Gandelman Milgrom

Yoav Menes

Yoli Shavit

Ido Shamay

Rotem Damsker

Methods, systems, and machine-readable mediums to perform a neural network to encode log data. In at least one embodiment, a processor comprising one or more circuits to encode at least one log message, at least in part, by encoding a first type of information in the at least one log message to obtain a first encoding, encoding a second type of information in the at least one log message to obtain a second encoding, and obtaining a resultant encoding at least in part by combing at least the first and second encodings.

Using neural networks to encode log data

Systems, computer program products, and methods are described herein for dynamic reconfiguration of network communications. An example system includes a first network pod including a first set of network ports, a second network pod including a second set of network ports, a set of network cores, and a first intermediate network switch. The first intermediate switch operatively couples the first network pod, the second network pod, and the set of network cores. The first intermediate network switch is configured to selectively establish full bisectional bandwidth data communication between a subset of the set of network cores, a subset of the first set of network ports, and a subset of the second set of network ports.

Systems and methods for dynamic reconfiguration of network communications

A networking device and system are described, among other things. An illustrative system is disclosed to include a switch programmed to route a received packet to an egress port based on a combination of a destination address associated with the received packet and an identification of an ingress port from which the packet was received by the switch.

Network partition filter

An apparatus, system, and method include, for each of two or more switches of a communication network, identifying a set of routing paths from the switch to a destination node based on a topology associated with the communication network. The set of routing paths include a first subset of routing paths and a second subset of routing paths. The topology includes an indication of a convergence of the first subset of routing paths at a node between the switch and the destination node. The apparatus, system, and method include allocating a data flow to a first routing path of the first subset of routing paths and a second routing path of the second subset of routing paths according to a target data flow rate common to the first routing path and the second routing path.

Adaptive routing for asymmetrical topologies

A routing controller () includes an interface () and multiple processors (). The interface is configured to receive a permutation () defining requested interconnections between N input ports and N output ports of a Benes network (). The Benes network includes multiple 2-by-2 switches (), and is reducible in a plurality of nested subnetworks associated with respective nesting levels, down to irreducible subnetworks including a single 2-by-2 switch. The multiple processors are configured to collectively determine a setting of the 2-by-2 switches that implements the received permutation, including determining sub-settings for two or more subnetworks of a given nesting level in parallel, and to configure the multiple 2-by-2 switches of the Benes network in accordance with the determined setting.

Efficient parallelized computation of a Benes network configuration

A network adapter includes a port and one or more circuits. The port is to send packets to a network in accordance with a Remote Direct Memory Access over Converged Ethernet (RoCE) protocol. The one or more circuits are to decide whether a packet is permitted to undergo Adaptive Routing (AR) in being routed through the network, to mark the packet with an indication of whether the packet is permitted to undergo AR, and to send the marked packet to the network via the port.

Marking of RDMA-over-converged-ethernet (RoCE) traffic eligible for adaptive routing

A method for distributed allocation of data paths in an optical network () including optical switches () connected by optical links (), includes receiving a request for a data path for connecting a source node () and a destination node (). In in response to the request, one or more queries are sent, the queries corresponding to one or more candidate optical circuits that connect the source node and the destination node, the queries requesting one or more processors () to configure the optical switches along the candidate optical circuits to reserve optical channels on the optical links of the candidate optical circuits for the requested data path. An optical circuit is identified from among the candidate optical circuits, in which all the optical channels for the requested data path have been reserved successfully. The requested data path is established over the identified optical circuit.

Distributed optical circuit allocation in optical data-center networks (ODCN)

Systems, computer program products, and methods are described herein for network discovery, port identification, and/or identifying fiber link failures in an optical network, in accordance with an embodiment of the invention. The present invention may be configured to sequentially connect each port of an optical switch to a network port of a server and generate, based on information associated with network devices connected to the ports, a network map. The network map may identify which network devices are connected to which ports of the optical switch and may permit dynamic port mapping for network installation, upgrades, repairs, and/or the like. The present invention may also be configured to determine a fiber link in which a failure occurred and reconfigure the optical switch to allow communication between an optical time-domain reflectometer and the fiber link to test the fiber link.

Systems and methods for identifying fiber link failures in an optical network

A method includes receiving, using a processing device, a first condition associated with an operation at a data center, where the operation at the data center pertains to a first location at the data center, the first location corresponding to a first parameter value. The method further includes providing the first condition as an input to a machine learning model. The method also includes performing one or more reinforcement learning techniques using the machine learning model to cause the machine learning model to output an indication of a final location associated with the operation, where the final location corresponds to a final parameter value that is closer to a target than the first parameter value corresponding to the first location at the data center.

Data center job scheduling using machine learning

A method includes providing a plurality of processes interconnected by a network, each of the plurality of processes being configured to hold a block of data destined for others of the plurality of processes. A set of data for all-to-all data exchange is received from one or more of the processes. The set of data is configured as a plurality of blocks of data in a matrix as matrix data, the matrix being distributed among the plurality of processes. The matrix data is transposed by changing the position of selected blocks of data of the plurality of blocks of data relative to the other blocks of data of the plurality of the blocks of data, without changing the structure of each of the blocks of data. The transposed matrix data is over the network and is then received, repacked, and conveyed to destination processes.

Collective communication system and methods

A network element () includes circuitry and at least one port (). The at least one port is coupled to an optical fabric () including one or more optical switches () that provide optical paths between the at least one port and multiple destination nodes, at predefined time slots. The circuitry is configured to hold a schedule plan () that specifies which of the destination nodes are accessible via the optical fabric at which of the time slots, to queue packets that are destined to the destination nodes, and to transmit the queued packets via the at least one port in accordance with the schedule plan.

Time division communication via optical switch fabric

A system for bandwidth estimation includes an interface and a processor. The interface communicates with a fat-tree (FT) network including multiple switches including (i) leaf switches belonging to a bottom level, (ii) spine switches belonging to a top level and (iii) intermediate switches belonging to one or more intermediate levels. Links connect between selected ones of the switches. The processor is to calculate, for a given level of the FT network that is divided into multiple groups of switches, oversubscription ratios for the respective groups, an oversubscription ratio of a group being indicative of a ratio between (i) a first available bandwidth on the links connecting the switches in the group to a lower level, and (ii) a second available bandwidth on the links connecting the switches in the group to a higher level, and to report a figure of merit of the FT network based on the oversubscription ratios.

Bandwidth evaluation in a fat-tree network

A routing controller () includes an interface () and multiple processors () The interface is configured to receive a permutation () defining requested interconnections between N input ports and N output ports of a Benes network (). The Benes network includes multiple 2-by-2 switches (), and is reducible in a plurality of nested subnetworks associated with respective nesting levels, down to irreducible subnetworks including a single 2-by-2 switch. The multiple processors are configured to collectively determine a setting of the 2-by-2 switches that implements the received permutation, including determining sub-settings for two or more subnetworks of a given nesting level in parallel, and to configure the multiple 2-by-2 switches of the Benes network in accordance with the determined setting.

A method for communication includes provisioning each node in a network with a respective set of two or more network addresses. Each node in succession is assigned a respective network address from the respective provisioned set that has not been assigned for use by any preceding node. Upon finding for a given node that all the network addresses in the respective provisioned set were assigned to preceding nodes, the preceding nodes are searched to identify a candidate node having an additional network address in the respective provisioned set, other than the assigned respective network address, that was not yet assigned to any of the nodes. The additional network address is assigned to the candidate node instead of the respective network address that was previously assigned to the candidate node, and the assigning of the network addresses to the nodes in the succession resumes following the candidate node.

Unique address assignment without reconfiguration

A method in which a plurality of process are configured to hold a block of data destined for other processes, with data repacking circuitry including receiving circuitry configured to receive at least one block of data from a source process of the plurality of processes, the repacking circuitry configured to repack received data in accordance with at least one destination process of the plurality of processes, and sending circuitry configured to send the repacked data to the at least one destination process of the plurality of processes, receiving a set of data for all-to-all data exchange, the set of data being configured as a matrix, the matrix being distributed among the plurality of processes, and transposing the data by each of the plurality of processes sending matrix data from the process to the repacking circuitry, and the repacking circuitry receiving, repacking, and sending the resulting matrix data to destination processes.

Growing community of inventors

Zichron Yaakov, Israel

Eitan Zahavi