Juniper Networks, Inc.

Cisco Technology, Inc.

Qingming Ma

Saravanan Agasaveeran

Krishna Narayanaswamy

Bryan Burns

Anupam Bharali

Zhanhe Shi

Xianzhi Li

Cheng-Lee Nee

Steven Howard Berl

Gopal K Dommety

Li Fei Zheng

Rohini Kasturi

Kannan Varadhan

Jiaxiang Su

Hongsheng Ping

Hancang Wang

Satish Gannu

Sudhakar R Padala

Perng-Hwa A Wu

John Gary Waclawsky

Frederick Juergens Baker

Sheng Shuang Li

Shan Yu

Na Liu

Jianhua Gu

Sanjay Gupta

Zhenyu Wang

Dongsheng Mu

Ricardo Oliveira

Xuejun Wu

Zeyong Lin

Vipin Rawat

Michael Chuong Shieh

A method includes a proxy device receiving from a source device a request to establish a flow to a destination device; generating, based on the request, a meta-packet that indicates that the flow to the destination device is to be proxied; determining whether a pre-established flow connecting the proxy device to another proxy device that leads toward the destination device exists; sending the meta-packet on the pre-established flow, when it is determined that the pre-established flow exists; receiving by the other proxy device, the meta-packet, and establishing the flow to the destination device based on the meta-packet, where the proxy devices assign one or more of a source address, a source port, a destination address, or a destination port, associated with the source device and the destination device, to the pre-established flow.

Flow control scheme for parallel flows

In general, techniques are described for network traffic pattern matching using adaptive deterministic finite automata (DFA). A network device may implement the techniques to promote pattern matching. The network device comprises a control unit that stores first and second data defining first and second portions of a DFA, respectively. The first data defines first states of the DFA in an uncompressed format. The second data defines second states of the DFA in a compressed format. The network device also includes an interface that receives network packets. The control unit processes the network packets to traverse the first and second states. The control unit then compares a number of times the first and second states have been traversed. Based on the comparison, the control unit dynamically reallocates the first states of the DFA in the uncompressed format and the second states of the DFA in the compressed format.

Network traffic pattern matching using adaptive deterministic finite automata

In one example, a network device includes a network interface that receives a packet, a storage card that stores session data for monitored network sessions, a plurality of service processing cards that process packets of respective subsets of the network sessions, wherein each of the service processing cards comprises a respective memory to store session data for the respective subset of the network sessions processed by the corresponding service processing card, and a switch fabric coupled to the network interface, the storage card, and the plurality of service processing cards. One or more of the plurality of service cards process the received packet based on the session data stored by the storage card. The one or more of the plurality of service cards retrieve the session data for the network session to which the packet corresponds from the storage card and store the retrieved session data in the respective memory.

Storing session information in network devices

A network device may implement packet scheduling with administrator-configurable packet scheduling policies. In one implementation, the network device includes a filter component configured to assign priority levels to data units, the priority levels defining traffic classes for the data units. The network device may also include a scheduler component configured to schedule transmission of the traffic classes based on an assignment of weights to the traffic classes using at least one bandwidth allocation policy that exhibits a bandwidth allocation profile that varies based on one or more parameters of the bandwidth allocation policy that are configurable by an administrator.

Weight-based bandwidth allocation for network traffic

In general, the invention is directed to techniques for improving memory utilization in a priority queuing system of a network device. More specifically, a priority queue memory management system is described in which memory pages are assigned to the various priority queues in order to implement an efficient first in, first out (FIFO) functionality. The dynamic memory techniques described herein allow the multiple priority queues to share a common memory space. As a result, each priority queue does not require a pre-allocated amount of memory that matches the aggregate size of the packets that must be buffered by the queue.

Memory utilization in a priority queuing system of a network device

A system determines a scheduling value based on a current length of a downstream queue in a network device. The system sends the scheduling value from the downstream queue to an upstream queue and schedules dequeuing of one or more data units, destined for the downstream queue, from the upstream queue based on the scheduling value.

Coordinated queuing between upstream and downstream queues in a network device

A data prefetching technique uses predefined prefetching criteria and prefetching models to identify and retrieve prefetched data. A prefetching model that defines data to be prefetched via a network may be stored. It may be determined whether prefetching initiation criteria have been satisfied. Data for prefetching may be identified based on the prefetching model when the prefetching initiation criteria have been satisfied. The identified data may be prefetched, via the network, based on the prefetching model.

Optimized prefetching for wide area networks

A method includes receiving a data unit, determining whether a current state, associated with a deterministic finite automata (DFA) that includes a portion of states in a bitmap and a remaining portion of states in a DFA table, is a bitmap state or not, and determining whether a value corresponding to the data unit is greater than a threshold value, when it is determined that the current state is not a bitmap state. The method further includes determining whether the current state is insensitive, when it is determined that the value corresponding to the data unit is greater than the threshold value, where insensitive means that each next state is a same state for the current state, and selecting a default state, as a next state for the current, when it is determined that the current state is insensitive.

Hybrid representation for deterministic finite automata

A method may include receiving a packet in a network device, selecting one of a group of ingress buffers, where each ingress buffer is associated with a different one of a group of processors, distributing the packet to the selected ingress buffer; and scheduling the packet, based on a congestion state of a queue in an egress buffer associated with the packet, to be processed by the processor associated with the selected ingress buffer to provide a network service.

Scalable resource management in distributed environment

A method is provided for queuing packets. A packet may be received and its flow identified. It may then be determined whether a flow queue has been assigned to the identified flow. The identified flow may be dynamically assigning to an available flow queue when it is determined that a flow queue has not been assigned to the identified flow. The packet may be enqueued into the available flow queue.

Flow-based queuing of network traffic

In general, the invention is directed to techniques of dynamically balancing network traffic load among multiple paths through a computer network. The techniques distribute and redistribute flows of network packets between different paths based on dynamically measured path bandwidth and loads of each flow. In distributing the flows, Quality of Service (QoS) bandwidth requirements of the flows may be maintained.

Dynamic flow-based multi-path load balancing with quality of service assurances

In general, the invention is directed to techniques of load balancing network traffic among paths on a point-to-multipoint overlay network. In load balancing the network traffic, the techniques take into consideration costs associated with paths through the overlay network and costs associated with paths beyond the egress points of the overlay network, even when such costs may be unequal.

Growing community of inventors

Santa Clara, CA, United States of America

Qingming Ma