Google Inc.

Junlan Zhou

Zhengrong Ji

Amin Vahdat

Yuguang Wu

Arjun Singh

Alexander Shraer

Jiangbo Li

Min Zhu

Shidong Zhang

Hong Liu

Ryohei Urata

Christopher Lyle Johnson

Ashish Naik

Joon Suan Ong

Rui Wang

Hassan Wassel

Leonid Poutievski

Shaileshh Bojja Venkatakrishnan

Keqiang He

Peter Peresini

Alok Kumar

Uday Ramakrishna Naik

Victor Lin

Sushant Jain

Steven Padgett

Glen Anderson

Philip Michael Wells

Tsuwei Chen

Ke Xiao

David Fermor

Ji Zhengrong

Methods are provided for mitigating hash correlation. In this regard, a hash correlation may be found between a first switch and a second switch in a network. In this network, a first egress port is to be selected among a first group of egress ports at the first switch for forwarding packets, and a second egress port is to be selected among a second group of egress ports at the second switch for forwarding packets, where the first group has a first group size and the second group has a second group size. Upon finding the hash correlation, a new second group size coprime to the first group size may be selected, and the second group of egress ports may be mapped to a mapped group having the new second group size. The second switch may be configured to route packets according to the mapped group.

Method to mitigate hash correlation in multi-path networks

Aspects of the present disclosure relate to synchronizing route updates. In one aspect, one or more computing devices, such as a router or a centralized controller, may receive a notification of a planned topology change to the network. The topology change may affect a link between nodes. Further, the one or more computing devices may determine one or more paths associated with the link and determine one or more user nodes associated with each path. The one or more computing devices may send an instruction message associated with the planned topology change to each user node of each path and receive an acknowledgment message in response to the instruction message. Based on the information in the acknowledgement message, the one or more computing devices may determine whether to effect the planned topology change.

System and method for synchronized route update

A system and method is provided for sending congestion notification messages through L3 networks and implementation of QCN in L3 switches. For example, according to this system and method, an L3 switch receives one or more data packets, and determines, based on the received one or more data packets, whether the L3 switch is congested. If the L3 switch is congested, it generates a congestion notification message, the congestion notification message including an Internet Protocol (IP) header, the IP header identifying a source of the one or more received data packets as its destination. The L3 switch sends the congestion notification message to the source of the one or more received data packets using information in the IP header.

System and method for providing congestion notification in layer 3 networks

Exemplary embodiments allocate network traffic among multiple paths in a network, which may include one or more preferred paths (e.g. shortest paths) and one or more alternative paths (e.g., non-shortest paths). In one embodiment, network traffic in form of flows may be allocated to the preferred paths until the allocation of additional network traffic would exceed a predetermined data rate. Additional flows may then be sent over the alternative paths, which may be longer than the preferred path. The paths to which each flow is assigned may be dynamically updated, and in some embodiments the path assignment for a particular flow may time out after a predetermined time. Accordingly, the flow traffic of each path may be balanced based on real-time traffic information.

Flow level dynamic load balancing

A system and method is provided for sending congestion notification messages through L3 networks. For example, a data packet is received at a first switch in a first fabric block of an L3 network, and the first switch performs source MAC tagging of the data packet. The data packet is then forwarded to a second switch in a second fabric block of the L3 network, and the source MAC tag is maintained by the second switch and any intermediate switches. The second switch determines, in response to receiving the data packet, whether it is congested, and generates a notification message if it is congested. The notification message is L2 forwarded to the first fabric block, and further forwarded from the first switch to a source of the data packet using ACL matching.

Overlapping flow rules are included in a ternary content addressable memory (TCAM), while still enabling a hardware counter to increment each of the overlapping rules when a packet matching each of the overlapping rules is transmitted through the TCAM. In a given set of flow specifications, a first flow specification is identified that overlaps with a second flow specification. Rules are determined corresponding to the first flow specification, the second flow specification, and an intersection of the first and second flow specifications. Priorities are assigned to each of the rules, wherein the rule corresponding to the intersection is assigned a higher priority than the rules corresponding to the first and second flow specifications. Such rules are stored in the TCAM.

System and method for measurement of flow statistics

Systems and methods for achieving high utilization of a network link are provided. A first communication protocol can be selected for transmitting network flows of a first type. A first quality of service can be assigned to network flows of the first type. A second communication protocol can be selected for transmitting network flows of a second type. A second quality of service, lower than the first quality of service, can be assigned to network flows of the second type. A first percentage of available bandwidth can be allocated to the network flows of both the first and second types. The remaining bandwidth, plus a second percentage of available bandwidth, can be allocated to the network flows of the second type, such that the total allocated bandwidth exceeds the available bandwidth of the network link.

Systems and methods for achieving high network link utilization

A router residing in a network comprises at least one ingress port, at least one egress port, and a processor programmed to compare at least two label switch paths, determine potential conflicts between the at least two label switch paths based on the ingress ports and egress ports utilized by the label switch paths, and determine a selected identifier to be assigned relative to each label switch path. The processor is configured to assign a common identifier if no conflict exists. A storage medium is operatively coupled to the processor for storing the selected identifiers related to the label switch paths. The processor may be configured to determine that a conflict exists between two label switch paths if they utilize the same ingress port on the router and different egress ports on the router.

System and method for improving routing labeling

The present disclosure presents a system and method for determining a logical topology of a network, given the network's physical topology. More particularly, a logical topology is implemented across a plurality of optical circuit switches that interconnect the nodes of a network. Each of the optical circuit switches includes an initial internal configuration. The internal configuration of the optical circuit switches are swapped to generate new logical topologies. A fitness is determined for each of the generated topologies. The fitnesses of the topologies are then ranked and the most fit logical topology is implemented in the network.

Randomized rotation striping for direct connect networks

The present technology considers multi-stage network topologies where it is not possible to evenly stripe uplinks from a lower stage of the network topology to switching units in an upper stage of the topology. This technology proposes techniques to both improve overall throughput and to deliver uniform performance to all end hosts with uneven connectivity among the different stages while delivering uniform performance to all hosts. To achieve improved network performance in case of asymmetric connectivity, more flows may be sent to some egress ports than to others, thus weighting some ports more than others, resulting in Weighted Cost Multi Path (WCMP) flow distribution.

Multi-stage switching topology

A technique for load balancing in a multi-topology network selects a network path having a favorable bandwidth*hop count product. Accordingly, shorter paths are prioritized while longer paths are used for traffic load balancing when appropriate. For example, a telecommunications network may employ a Clos-style topology in which hosts are connected to each other through multiple hierarchical levels of forwarding devices. One or more high level switches may be removed, creating isolated blocks of forwarding devices. The blocks may be connected by the remaining original high level Clos-style topology switch(es) and a second type of network topology. In an exemplary embodiment, traffic may be scheduled over routes with the following decreasing priorities: (1) the shortest path, using either or both network topologies; (2) a path using only a single network topology type; (3) non-shortest paths between directly-connected blocks; and (4) non-shortest paths between indirectly-connected blocks.

Traffic load balancing in a multi-connect topology

Exemplary embodiments provide changes to routing schemes, i.e. WCMP groups or WCMP sets, installed in a network traffic distribution table, e.g. multipath table. WCMP groups of a multipath table are updated to accommodate a new WCMP group. This can be achieved by reducing the size of the existing WCMP groups on the multipath table. The goal is to reduce the existing WCMP groups just enough to make room for the new WCMP group. An objective is to minimize the number of existing WCMP groups to be reduced before a new WCMP group can be installed in the multipath table.

Efficient updates of weighted cost multipath (WCMP) groups

Systems and methods for increasing bandwidth in a computer network are provided. A computer network can include a first lower level switch, first and second upper level switches, and first and second passive optical splitters, and a mirror. The first passive optical splitter can have a first port directly coupled to the first upper level switch, a second port directly coupled to the second upper level switch. The second passive optical splitter can have a port directly coupled to a port of the first passive optical splitter, and a port directly coupled to the first lower level switch. The mirror can be coupled to a port of the second passive optical splitter and reflect an optical signal received from the second passive optical splitter to the first upper level switch and second upper level switch through the second passive optical splitter and the first passive optical splitter.

Growing community of inventors

Sunnyvale, CA, United States of America

Junlan Zhou