Google Inc.

Ashish Naik

Amin Vahdat

Arjun Singh

Subhasree Mandal

Leon Poutievski

Junlan Zhou

Milad Olia Hashemi

Gufeng Zhang

Victor Lin

Philip Michael Wells

Jiangbo Li

Ke Xiao

Shidong Zhang

Methods, systems and apparatus, including computer programs encoded on computer storage medium, for predicting a likelihood of a future computer memory failure. In one aspect training data inputs are obtained, where each training data input includes correctable memory error data that describes correctable errors that occurred in a computer memory and data indicating whether the correctable errors produced a failure of the computer memory. For each training data input, image representations of the correctable memory error data included in the training data input are generated. The image representations are processed using a machine learning model to output an estimated likelihood of a future failure of the computer memory. A difference between the estimated likelihood of the future failure of the computer memory and the data indicating whether the correctable errors produced a failure of the computer memory is computed. Values of model parameters are updated using the computed difference.

Image models to predict memory failures in computing systems

A system and method are provided for updating a network including at least one optical circuit switch (OCS) to transition from an existing network topology to a new network topology. One or more intermediate topologies between the existing topology and the new topology are created. Creating the intermediate topologies includes selecting first links to be added to the existing topology without removing links, selecting additional links to be added to the existing topology upon removal of one or more existing links, and adding one or more of the selected first and additional links to the existing topology to create a first intermediate topology. It is determined whether any of the selected first and additional links are still to be added, and if no selected first and additional links are to be added, remaining links are removed. The transition from the existing topology to the first intermediate topology is then effected.

System and method for adjusting network topology without packet loss

System and methods for efficiently distributing data packets in a multi-path network while maintaining flow affinity are provided. In one aspect, a system and method includes calculating hash values for distributing different flows, or sets of flows, of data packets received at a routing device. The hash value is calculated not only using information in the data packets, but also based on additional information that is determined based on an N-bit derived from the data packets. In some cases, calculating a hash value based on the additional information increases the entropy of the hashing function, thus enabling a routing device to distribute different flows of data packets over a greater number of network paths. Each routing device can derive a different, and randomly generated N-bit key while maintaining flow affinity for each received data packet in a given flow of data packets.

Traffic distribution over multiple paths in a network while maintaining flow affinity

System and methods for efficiently distributing data packets in a multi-path network having a plurality of equal cost paths each having the same number of hops are provided. In one aspect, a system and method includes reordering the entries in the routing table of one or more peer routing devices, such that succeeding equal cost next hop routing devices that are interconnected in a Clos formation to the peer routing devices received unbiased traffic from each of the peer routing devices, thus enabling the succeeding next hop routing devices to distribute data traffic evenly over additional equal cost multiple hops further into the network.

Traffic distribution over multiple paths in a network

Aspects of the present disclosure relate to detecting and repairing permanently pauses on a flow controlled fabric. In one aspect, one or more computing devices, such as a switch or a centralized controller, may detect whether a port of a network device receives one or more pause messages. The pause messages may instruct the network device to pause data transmission. Further, the one or more computing devices may determine a period of time during which the port receives the one or more pause messages and identify the port as a permanently paused port based on the determined period of time. The one or more computing devices may then reconfigure the permanently paused port to stop complying with the one or more pause messages.

Detection and repair of permanent pause on flow controlled fabric

Network performance may be improved by adjusting a logical network topology while avoiding attendant disruptions. By identifying an alternative logical topology, routing network traffic away from links that will be eliminated from the alternative logical topology, and updating a traffic routing plan upon the adoption of the alternative logical topology, a network may adopt a more efficient topology without misdirecting network traffic.

Networking systems with dynamically changing topologies

A patch panel device has communication ports respectively associated with a unique binary number having a number of binary digits, and a plurality of computing devices are each respectively coupled to at least one of the ports. A connectivity test includes configuring the patch panel device to send a message to a first sending device in response to receiving data at a port associated with a binary number having a first value as the respective binary digit, and not to send a message to a second sending device in response to receiving data at a port associated with a binary number having a second value as the respective binary digit. Each of the computing devices is caused to transmit data for arrival at a corresponding port, an indication is received for each device regarding whether the device received a message. A connectivity map is generated.

Determining physical connectivity of data center devices

A method and apparatus are provided for communicating congestions that area detected in a communications network. According to the method and apparatus, a first Ethernet frame encapsulating a first PDU is transmitted across an Ethernet network. Congestion in the network is then detected, and an indication of the congestion is inserted in a predetermined header field of the first Ethernet frame. Once the first Ethernet frame is received by an endpoint in the network, the indication of congestion is transferred into a second PDU that is associated with the first PDU. The second PDU is then encapsulated in a second Ethernet frame and transmitted towards the sender of the first PDU.

Growing community of inventors

Los Altos, CA, United States of America

Ashish Naik