Oracle America, Inc.

Sun Microsystems, Inc.

Google Inc.

Future Wei Technologies, Inc.

Sanmina Corporation

Oracle America, Inc. Formerly Known As Sun Microsystems, Inc.

Kais Belgaied

Nicolas G Droux

Sunay Tripathi

Darrin Paul Johnson

Erik Nordmark

Xiangyu Dong

Pedram Zare Dashti

Paul Sangster

Darren J Moffat

Jianqiao Liu

Bhargava K Yenduri

Gary W Winiger

Mark C Powers

Franz Michael Schuette

Anders D Persson

Cesar A C Marcondes

Tim P Marsland

Eric T Cheng

Rajagopal Kunhappan

Hua Ji

Yazhou Zu

Darren J Reed

Sumedha K Swamy

Aleksandr Guzovskiy

Richard Elling

Nitin Hande

Yunsong Lu

This disclosure generally provides solutions for improving the performance of a custom-built, packet-switched, TPU accelerator-side communication network. Specifically a set of solutions to improve the flow-control behavior by tuning the packet buffer queues in the on-chip router in the distributed training supercomputer network are described.

Queue allocation in machine learning accelerators

A uniform and unified firmware in-field upgrade capability for the optics modules may ensure compatibility, security and code quality, and scalability. In some examples, an intermediate representation, which includes vendor firmware upgrade operations and control logic, may be defined, received, and parsed. Read/write operations may be communicated to optical module(s) based on the control logic. In some examples, a unified optics module firmware in-field upgrade framework, which has multiple defined software layers, may ensure a uniform and unified approach to managing optics module(s) from different vendors and used by different projects. The software layers that may properly translate optics module read/write operations, abstract and make uniform the read/write operations, provide libraries of intermediate representations, package the intermediate representations into executables/scripts, monitor optics module status, determine when a new firmware is released, and gradually upgrade the optics module firmware.

Methods for optics module firmware in-field upgrade

A server computer is configured to write a first copy of a block of data to a first namespace on a first non-volatile memory-based cache drive and a second copy of the block of data to a RAID controller for de-staging of the data to hard disk drives of a RAID array. Acknowledgment of hardening of the data on the hard disk drives initiates purging of the first copy of the block of data from the cache drive. High availability is enabled by writing a third copy of the block of data to a second server to store the block of data in a second namespace on a second non-volatile memory-based cache drive. Restoring of data after power loss accesses the data on the first non-volatile memory-based cache drive.

Method and apparatus for consistent and highly available data storage using local and fabric attached non-volatile memory storage devices

A method for changing network configuration parameters that includes generating a request to change a network configuration parameter, where the request is generated by a virtual machine, sending the request to a virtual network interface card (VNIC) associated with the virtual machine, sending the request to a VNIC configuration database associated with the VNIC, determining whether the virtual machine is allowed to change the network configuration parameter, if the virtual machine is allowed to change the network configuration parameter, updating the VNIC configuration database and VNIC to reflect the change in the network configuration parameter, and notifying the virtual machine that the change in network configuration parameter is allowed, and if the virtual machine is not allowed to change the network configuration parameter, dropping the request.

Method and system for network configuration for virtual machines

A system and method for providing network connectivity to a host, involving creating a virtual switch on the host, specifying at least one data link attribute of the virtual switch, creating a plurality of virtual network interface cards (VNICs) on the host, associating each of the plurality of VNICs with the virtual switch, and assigning the at least one data link attribute of the virtual switch to each of the plurality of VNICs, where the virtual switch is connected to a physical network interface card (NIC) associated with the host, where each of the plurality of VNICs is associated with a different one of a plurality of execution environments, where the plurality of execution environments is located on the host, and where the plurality of VNICs is located on the host.

Enhanced virtual switch

A method for notifying a packet destination that includes receiving a packet by a network interface card (NIC), where the packet destination is a destination of the packet, classifying the packet, forwarding the packet to one of a plurality of receive rings on the NIC, determining whether the one of the plurality of receive rings comprises space to store the packet, dropping the packet if the receive ring does not comprise the space to store the packet, and sending a notification message to the packet destination, where the notification message indicates that the packet was dropped by the receive ring.

Notifying network applications of receive overflow conditions

A system and method for massively multi-core computing are provided. A method for computer management includes determining if there is a need to allocate at least one first resource to a first plane. If there is a need to allocate at least one first resource, the at least one first resource is selected from a resource pool based on a set of rules and allocated to the first plane. If there is not a need to allocate at least one first resource, it is determined if there is a need to de-allocate at least one second resource from a second plane. If there is a need to de-allocate at least one second resource, the at least one second resource is de-allocated. The first plane includes a control plane and/or a data plane and the second plane includes the control plane and/or the data plane. The resources are unchanged if there is not a need to allocate at least one first resource and if there is not a need to de-allocate at least one second resource.

System and method for massively multi-core computing systems

In general, the invention relates to a method for processing packets. The method includes receiving a first packet by a network interface card (NIC) connected to a host, classifying the first packet using a classifier, sending the first packet to a receive ring based on a classification of the first packet by the classifier, and sending the first packet from the receive ring to a first virtual network interface card (VNIC) located on the host. The method further includes determining, using a first policy associated with the first VNIC, whether to process the first packet using offload hardware. When the first packet is to be processed using the offload hardware, the method includes sending the first packet to the offload hardware, receiving a first processed packet from the offload hardware by the first VNIC and sending the first processed packet from the first VNIC to a first packet destination.

Method and system for onloading network services

A method for configuring a network on a host includes obtaining a first virtual network stack and a second virtual network stack on the host, configuring a first transport layer implementation on the first virtual network stack, configuring a second transport layer implementation on the second virtual network stack, receiving a packet by the host, sending a packet to the first virtual network stack, and processing the packet using the first transport layer implementation.

Network stack instance architecture with selection of transport layers

A method for processing packets. The method includes receiving a first packet by a network interface card (NIC) from a network, determining, using a first classification level, a first receive ring group (RRG) for the first packet, determining, using a second level classification, a first receive ring (RR) in the first RRG for the first packet, sending the first packet to the first RR, and sending the first packet from the first RR to a host operatively connected to the network interface card, wherein the first packet is received by a first virtual network interface card (VNIC) associated with the first RRG, where the first RRG is located in the NIC.

Method and system for classifying packets in a network interface card and interface for performing the same

A method for migrating a virtual machine executing on a host. The method involves monitoring, by a monitoring agent connected to a device driver, hosts in a network, wherein the device driver is connected to a network interface card, determining a virtual machine to be migrated based on a virtual machine policy, sending, by the host, a request to migrate to at least one of a plurality of target hosts in the network, receiving an acceptance to the request to migrate from at least one of the plurality of target hosts, determining, by the monitoring agent, a chosen target host to receive the virtual machine based on a migration policy, wherein the chosen target host is one of the at least one target hosts that sent the acceptance, sending a confirmation and historical information to the chosen target host, and migrating the virtual machine to the chosen target host.

System for migrating a virtual machine and resource usage data to a chosen target host based on a migration policy

A method for verifying an Executable and Linking File (ELF) object, that includes receiving a request for an ELF object from a client, obtaining the ELF object, determining whether a signature associated with the ELF object is valid, determining whether a usage restriction is associated with the ELF object, if the signature is valid, and restricting access to the ELF object, if the usage restriction is associated with the ELF object.

System and computer readable medium for verifying access to signed ELF objects

In general, the invention relates to a creating a network model on a host. The invention includes: gathering first component properties associated with a first physical network device on a target network; creating a first container using first component properties; determining that a second physical network device is operatively connected to the first physical network device via a physical network link; gathering second component properties associated with the physical network link; creating a first VNIC associated with the first container; determining that at least one virtual network device is executing on the second physical network device; gathering third component properties associated with the at least one virtual network device; creating a second container, wherein the second container is configured using the third component properties; and creating a second VNIC associated with the second container.

Growing community of inventors

Sunnyvale, CA, United States of America

Kais Belgaied