Vmware, Inc.

Wmware, Inc.

Lenin Singaravelu

Shilpi Agarwal

Haoqiang Zheng

Garrett Cale Smith

Ayyappan Veeraiyan

Daniel Michael Hecht

Yong Wang

Jin Heo

Jui-Ting Weng

Chien-Chia Chen

Amitabha Banerjee

Ruijin Zhou

Xiaobo Huang

Rishi Mehta

Debessay Fesehaye Kassa

A method of optimizing network processing in a system comprising a physical host and a set of physical network interface controllers (PNICs) is provided. The physical host includes a forwarding element. The method includes determining that a set of conditions is satisfied to bypass the forwarding element for exchanging packets between a particular data compute node (DCN) and a particular PNIC. The set of conditions includes the particular DCN being the only DCN connected to the forwarding element and the particular PNIC being the only PNIC connected to the forwarding element. The method exchanges packets between the particular DCN and the particular PNIC bypassing the forwarding element. The method determines that at least one condition in said set of conditions is not satisfied. The method utilizes the forwarding element to exchange packets between the particular DCN and the particular PNIC.

Specializing virtual network device processing to avoid interrupt processing for high packet rate applications

A host computer has a plurality of containers including a first container executing therein, where the host also includes a physical network interface controller (NIC). A packet handling interrupt is detected upon receipt of a first data packet associated with the first container. If the first virtual machine is latency sensitive, then the packet handling interrupt is processed. If the first virtual machine is not latency sensitive, then the first data packet is queued and and processing of the packet handling interrupt is delayed.

Networking stack of virtualization software configured to support latency sensitive virtual machines

In order to facilitate efficient and scalable lookup of current hop limits of transmitted packets, a communications device embeds hop limit values along with other connection parameters in a connection data structure. To transmit a packet for a particular connection, the communications device retrieves the data structure for the particular connection and applies the hop limit value embedded in the data structure to the packet for transmission. To keep track of the hop limits being embedded in different data structures of different connections, the communications device use a binary search in which each node of the search tree correspond to a different connection. The communications device maintains one such search tree per communications interface.

Fast lookup and update of current hop limit

Certain embodiments described herein are generally directed to improving performance of one or more machines within a system by clustering multidimensional datasets relating to the performance of the machines using inter-group dissimilarities between groups of the dataset. The method for improving performance of one or more machines within a system, includes forming a multidimensional dataset having a plurality of groups using performance related data associated with one or more machines in the system, clustering the plurality of groups into one or more clusters based on intergroup dissimilarities between the plurality of groups, identifying one or more anomalous clusters from among the one or more clusters, identifying the one or more anomalous groups in the one or more anomalous clusters, and adjusting a configuration of the system to improve the performance of the one or more machines corresponding to the one or more anomalous groups.

Group clustering using inter-group dissimilarities

Exemplary methods, apparatuses, and systems include receiving time series data for each of a plurality of performance metrics. The time series data is sorted into buckets based upon an amount of variation of time series data values for each performance metric. The time series data in each bucket is divided into first and second clusters of time series data points. The bucket having the greatest distance between clusters is used to determine a performance metric having a greatest distance between clusters. The performance metric having the greatest distance between clusters is reported as a potential root cause of a performance issue.

Detecting and remediating root causes of performance issues

Specializing virtual network device processing to bypass forwarding elements for high packet rate applications

A host computer has a virtualization software that supports execution of a plurality of virtual machines, where the virtualization software includes a virtual machine monitor for each of the virtual machines, and where each virtual machine monitor emulates a virtual central processing unit (CPU) for a corresponding virtual machine. A virtual machine monitor halts execution of a virtual CPU of a virtual machine by receiving a first halt instruction from a corresponding virtual machine and determining whether the virtual machine is latency sensitive. If the virtual machine is latency sensitive, then a second halt instruction is issued from the virtual machine monitor to halt a physical CPU on which the virtual CPU executes. If the virtual machine is not latency sensitive, then a system call to a kernel executing on the host computer is executed to indicate to the kernel that the virtual CPU is in an idle state.

Virtual machine monitor configured to support latency sensitive virtual machines

A host computer has one or more physical central processing units (CPUs) that support the execution of a plurality of containers, where the containers each include one or more processes. Each process of a container is assigned to execute exclusively on a corresponding physical CPU when the corresponding container is determined to be latency sensitive. The assignment of a process to execute exclusively on a corresponding physical CPU includes the migration of tasks from the corresponding physical CPU to one or more other physical CPUs of the host system, and the directing of task and interrupt processing to the one or more other physical CPUs. Tasks of of the process corresponding to the container are then executed on the corresponding physical CPU.

CPU scheduler configured to support latency sensitive virtual machines

Some embodiments provide a queue management system that efficiently and dynamically manages multiple queues that process traffic to and from multiple virtual machines (VMs) executing on a host. This system manages the queues by (1) breaking up the queues into different priority pools with the higher priority pools reserved for particular types of traffic or VM (e.g., traffic for VMs that need low latency), (2) dynamically adjusting the number of queues in each pool (i.e., dynamically adjusting the size of the pools), (3) dynamically reassigning a VM to a new queue based on one or more optimization criteria (e.g., criteria relating to the underutilization or overutilization of the queue).

Traffic and load aware dynamic queue management

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 tuning engine for a virtualized computing system is described that periodically collect performance metrics from the virtualized computing system, and detects whether a change in system state has occurred based on the collected metrics. The tuning engine may determine whether the virtualized computing system is densely virtualized, and accordingly modify operations and configuration settings of various components in charge of handling networking for the virtualized computing system.

Growing community of inventors

Sunnyvale, CA, United States of America

Lenin Singaravelu