Nicira, Inc.

Vmware, Inc.

Aayush Saxena

Pankaj Thakkar

Igor Ganichev

Yang Ping

Aravinda Kidambi Srinivasan

Teemu Koponen

Jianjun Shen

Dexiang Wang

Feng Pan

Amit Chopra

Chen Li

Bin Qian

Alexander Yip

Ganesan Chandrashekhar

Anuprem Chalvadi

Da Wan

Harold Vinson C Lim

Shekhar Chandrashekhar

Some embodiments provide a method for a network controller that manages multiple managed forwarding elements (MFEs) that implement multiple logical networks. The method stores (i) a first data structure including an entry for each logical entity in a desired state of the multiple logical networks and (ii) a second data structure including an entry for each logical entity referred to by an update for at least one MFE. Upon receiving updates specifying modifications to the logical entities, the method adds separate updates to separate queues for the MFEs that require the update. The separate updates reference the logical entity entries in the second data structure. When the second data structure reaches a threshold size in comparison to the first data structure, the method compacts the updates in at least one of the queues so that each queue has no more than one update referencing a particular logical entity entry.

Management of update queues for network controller

Certain embodiments described herein are generally directed to handling a hypervisor restart event in a distributed network system. Embodiments include receiving, by a central controller, a session identifier from a first hypervisor. Embodiments further include comparing, by the central controller, the session identifier to a stored session identifier associated with the first hypervisor. Embodiments further include determining, by the central controller based on the session identifier not matching the stored session identifier associated with the first hypervisor, that the first hypervisor has restarted. Embodiments further include updating, by the central controller, the stored session identifier associated with the first hypervisor to match the session identifier. Embodiments further include identifying, by the central controller, a second hypervisor that is associated with the first hypervisor. Embodiments further include sending, by the central controller, a notification to the second hypervisor that the first hypervisor has restarted.

Hypervisor restart handling mechanism using random session numbers

Certain embodiments described herein are generally directed to techniques for computing grouping object memberships in a network. Embodiments include receiving a plurality of network configuration updates. Embodiments include identifying delta updates to a plurality of grouping objects based on the plurality of configuration updates. Embodiments include determining a parallel processing arrangement for the delta updates based on dependencies in a directed graph comprising representations of the plurality of grouping objects. Embodiments include processing the delta updates according to the parallel processing arrangement in order to determine memberships of the plurality of grouping objects. Embodiments include distributing one or more updates to one or more endpoints based on the memberships of the plurality of grouping objects.

Differential dataflow approach for computing grouping membership in network controllers

Some embodiments provide a method for a first network controller executing at a first site of multiple sites spanned by a logical network. Network controllers execute at each site. The method generates logical network state data for the first site based on (i) data received from computing devices that implement the logical network at the first site and (ii) logical network configuration data from a network manager at the first site. The method provides the logical network state data for the first site to a second network controller executing at a second site. The method provides logical network state data received from the second site to the computing devices that implement the logical network at the first site.

Network controller for multi-site logical network

Certain embodiments described herein are generally directed to determining the spans of logical entities in a logical network using a graph theoretic method. For example, in some embodiments, a configuration of the logical network is represented as a directed graph with labeled edges. The span of any logical entity in the logical network may then be computed based on the directed graph of the logical network. For example, for a given entity represented by a given node, the set of nodes that can be reached from the given node via span-depends-on edges only may be computed and referred to as the span base of the given node. Further, the set of all managed forwarding elements reachable from each node of the span base of the given node via contains edges only is the span for the given entity.

Logical network configuration span

Certain embodiments described herein are generally directed to determining efficient update schedules for logical networks based on a graph representing the priority of features supported by the network, and a graph representing the priority of logical entities within the logical network. By building a graph of the logical network using directed edges to represent the update priority of one logical entity over another, an ordering, or a set of possible orderings, may be obtained. A topological sort of the graph may obtain such an ordering. Alternative orderings may indicate opportunities for parallel updating, which may result in a more efficient process for updating the logical network.

Priority based scheduling in network controller using graph theoretic method

Growing community of inventors

San Jose, CA, United States of America

Aayush Saxena