Vmware, Inc.

Telefonaktiebolaget Lm Ericsson (publ)

Nicira, Inc.

Ericsson Ab

Other

Tellabs San Jose, Inc.

Rishi Mehta

Sriganesh Kini

Boon Seong Ang

Wenyi Jiang

Rajeev Nair

Aditya Gajanana Holla

Amitabha Banerjee

Xiaochuan Shen

Banit Agrawal

Ayyappan Veeraiyan

Pritam Baruah

Ramya Bolla

Neelima Balakrishnan

Siming Li

Ninad Ghodke

Jayant Jain

Ganesan Chandrashekhar

Dexiang Wang

Guolin Yang

Lenin Singaravelu

Petr Vandrovec

Peter Arberg

Xunjia Lu

Bin Qian

Kalyan Saladi

Sanal Pillai

Sanjay Khanna

Seongbeom Kim

Ravindra Sunkad

Shrikrishna Khare

David S Curry

Samer Isam Nubani

Aditya Krishna Sonthy

Sujatha Sundararaman

Robert Colvin

Jinqiang Yang

Man-Tung T Hsiao

Paul M Hallinan

Naga S S Kishore Kankipati

Some embodiments provide a method for scheduling networking threads associated with a data compute node (DCN) executing at a host computer. When a virtual networking device is instantiated for the DCN, the method assigns the virtual networking device to a particular non-uniform memory access (NUMA) node of multiple NUMA nodes associated with the DCN. Based on the assignment of the virtual networking device to the particular NUMA node, the method assigns networking threads associated with the DCN to the same particular NUMA node and provides information to the DCN regarding the particular NUMA node in order for the DCN to assign a thread associated with an application executing on the DCN to the same particular NUMA node.

Optimized networking thread assignment

A method for managing several queues of a network interface card (NIC) of a computer. The method initially configures the NIC to direct data messages received for a data compute node (DCN) executing on the computer to a default first NIC queue. When the DCN requests data messages addressed to the particular DCN to be processed with a first feature for load balancing data messages across multiple queues and a second feature for aggregating multiple related data messages into a single data message, the method configures the NIC to direct subsequent data messages received for the DCN to a second queue in a first subset of queues associated with the first feature if a load on the default first queue exceeds a first threshold. Otherwise, if a load on the first subset of queues exceeds a second threshold, the method configures the NIC to direct subsequent data messages received for the particular DCN to a third queue in a second subset of queues associated with both the first and second features.

Dynamic processing of packets using multiple receive queue features

Some embodiments provide a method for presenting packets captured in a network. The method identifies a first set of packets from a first packet group of multiple captured packet groups. The method identifies a second set of packets, that corresponds to the first set of packets, from a second packet group of the multiple captured packet groups. The method displays representations of the multiple captured packet groups. At least one of the first set of packets and the second set of packets are presented with a different appearance from other packets of their respective packet group.

Analysis of simultaneous multi-point packet capture and display of the analysis

Computer system and method for characterizing throughput performance of a datacenter utilize bandwidth information of physical network interfaces in the datacenter and results of benchmark testing on throughput on a single processor core to compute a plurality of throughput constraints that define a throughput capacity region for the datacenter to improve throughput performance of the datacenter.

Performance characterization for datacenters

Some embodiments provide a method for managing multiple queues of a network interface card (NIC) of a host computer that executes a data compute node (DCN). The method defines first, second, and third subsets of the queues. The first subset of queues is associated with a first feature for processing data messages received by the NIC, the second subset of queues is associated with a second feature, and the third subset is associated with both features. The method receives a request from the DCN to process data messages addressed to the DCN using both the first and second features. The method configures the NIC to direct data messages received for the DCN to a queue that is selected from the third subset of queues.

An approach for dynamically distributing RSS engines to virtual machines based on flow data is disclosed. A method comprises receiving first absolute counts of data packets that belong to at least one data flow. Flow load indicator values are computed based on the first absolute counts, and stored in a lookup table. A sorted table is generated by sorting entries of the lookup table. A first count of filters that can be applied on RSS engines is determined. A second count of data flows in the sorted table and having corresponding flow load indicator values exceeding a threshold value is determined. If the second count exceeds the first count, then the first count of data flows is selected from the sorted table. The first count of filters that correspond to the data flows is determined, and the first count of the filters is assigned to at least one RSS engine.

Dynamic distribution of RSS engines to virtual machines based on flow data

In a computer-implemented method for reducing delay of bursty data transmission in a network employing a congestion control protocol, data is accessed that is to be periodically transmitted over a network employing a congestion control protocol. The data is to be periodically transmitted with a high burst rate followed by an idle period. The congestion control protocol progressively increases a data transmission rate during a data transmission rate increase period invoked immediately following a predetermined idle period. Prior to transmitting the data, priming data is transmitted during at least a portion of the idle period until the congestion control protocol progressively increases the data transmission rate to a desired transmission rate. The data is transmitted at the desired transmission rate.

Bursty data transmission in a congestion controlled network

Systems and methods described herein align various types of hypervisor threads with a non-uniform memory access (NUMA) client of a virtual machine (VM) that is driving I/O transactions from an application so that no remote memory access is required and the I/O transactions can be completed with local accesses to CPUs, caches, and the I/O devices of a same NUMA node of a hardware NUMA system. First, hypervisor of the VM detects whether the VM runs on a single or multiple NUMA nodes. If the VM runs on multiple NUMA nodes, a NUMA client on which the application is executing the I/O transactions is identified and knowledge of resource sharing between the NUMA client and its related hypervisor threads is established. Such knowledge is then utilized to schedule the NUMA client and its related hypervisor threads to the same NUMA node of the NUMA system.

Systems and methods for non-uniform memory access aligned I/O for virtual machines

Example methods are provided for a host to perform large receive offload (LRO) processing in a virtualized computing environment. The method may comprise receiving, via a physical network interface controller (NIC), incoming packets that are destined for the virtualized computing instance, and processing the incoming packets to generate at least one processed packet using a networking service pipeline that includes a packet aggregation service and multiple networking services. The packet aggregation service may be configured to aggregate the incoming packets into an aggregated packet and enabled at a service point along the networking service pipeline based on an LRO capability of at least one of the multiple networking services to process the aggregated packet. The method may also comprise forwarding the at least one processed packet generated by the networking service pipeline to the virtualized computing instance.

Large receive offload (LRO) processing in virtualized computing environments

Some embodiments provide a method for performing a multi-point capture of packets in a network. The method identifies multiple nodes for the multi-point capture in the network. The method configures each node of the multiple nodes to capture a set of packets. The method receives multiple captured packet sets from the multiple nodes. The method analyzes the multiple captured packet sets.

Distributed network troubleshooting using simultaneous multi-point packet capture

Techniques disclosed herein provide an approach for using receive side scaling (RSS) offloads from a physical network interface controller (PNIC) to improve the performance of a virtual network interface controller (VNIC). In one embodiment, the PNIC is configured to write hash values it computes for RSS purposes to packets themselves. The VNIC then reads the hash values from the packets and places the packets into VNIC RSS queues, which are processed by respective CPUs, based on the hash values. CPU overhead is thereby reduced, as RSS processing by the VNIC no longer requires computing hash values. In another embodiment in which the number of PNIC RSS queues and VNIC RSS queues are identical, the VNIC may map packets from PNIC RSS queues to VNIC RSS queues using the PNIC RSS queue ID numbers, which also does not require the computing RSS hash values.

Virtual network interface controller performance using physical network interface controller receive side scaling offloads

A method for load balancing IPTV channels is described. In one embodiment of the invention, a first Provider Edge (PE) network element of a label switched network, coupled with a second PE network element over multiple member links of an aggregate link, receives IPTV packets. For each IPTV packet received, the first PE network determines layer 3 information of the IPTV packet, and generates one or more channel load balancing keys based on the layer 3 information. The PE network element generates a hash value from the channel load balancing keys and determines which one of multiple member links to transmit the IPTV packet on based on the hash value, and transmits the IPTV packet to the second PE network element on the determined member link. Other methods and apparatuses are also described.

Load balancing pseudowire encapsulated IPTV channels over aggregated links

Systems and methods for preferentially assigning virtual machines (VMs) on a particular NUMA node with network queues on the same NUMA node are described. A load balancer process on a host assigns multiple VMs to network queues. The assignment of the VMs to a network queues is performed with a bias toward assigning VMs using a particular NUMA node to network queues on the same NUMA node. A scheduler on the host assigns VMs to NUMA nodes. The scheduler is biased toward assigning VMs to the same NUMA node as the PNIC and/or the same NUMA node as a network queue assigned to the VM.

NUMA I/O aware network queue assignments

A first network element that is to be coupled with a second network element over a first network in the same domain is described. The first network element includes a connectivity check detection module and a connectivity check fault state change module. The connectivity check detection module detects, according to a first protocol, a connectivity status of a local maintenance endpoint on a second network changing between not active and active. The connectivity check fault state change module creates and sends a first fault state change message, according to a second protocol, in response to the connectivity status of the local maintenance endpoint changing from not active to active instead of forwarding connectivity check messages periodically received by the first network element to a remote maintenance endpoint coupled to the second network element over a third network.

Scalable connectivity fault management in a bridged/virtual private LAN service environment

Growing community of inventors

San Jose, CA, United States of America

Rishi Mehta