Microsoft Technology Licensing, LLC

Omar Cardona

Alireza Dabagh

Khoa Anh To

Dmitry Malloy

Gabriel Silva

Donald Stanwyck

Alexander Malysh

Gerardo Diaz-Cuellar

Dinesh Kumar Govindasamy

Jacob Kappeler Oshins

John Andrew Starks

Daniel Max Firestone

Harish Srinivasan

Osman Nuri Ertugay

Madhan Sivakumar

Keith Edgar Horton

Gregory M Cusanza

Shankar Seal

Nicholas David Wood

Craig Daniel Wilhite

Poornananda Gaddehosur Ramachandra

Sze K Li

Jason Messer

Thomas Edward Molenhouse

Kamran Reypour

Anirban Paul

Satish Goswami

Chenyan Liu

Tom Xanthos

Examples of the present disclosure describe systems and methods for transmit side scaling. In examples, transmission side configuration information is received by a host operating system from a guest operating system, where the transmission side scaling configuration information specifies the manner in which data packets transmitted by the host operating system are to be distributed to a network interface card of the host operating system. The transmission side configuration scaling information is implemented in an outbound transmission table of the host operating system. When a data packet is received by the host operating system, the host operating system evaluates the data packet using the outbound transmission table. Based on the evaluation, that data packet is transmitted using a specified transmit queue of the network interface card.

Transmit side scaling and alignment

Techniques for aggregating execution metrics during virtualization are provided. In some embodiments, aggregated execution metrics (e.g., average execution time) are generated and stored for different types of supported virtualization service operations executed by a virtualization service provider (VSP) in a virtualization stack handling requests from a virtualization service client (VSC) running in a computer system emulator. For example, execution calls to the VSP are intercepted, and execution metrics for a triggered virtualization service operation are generated and aggregated into an aggregation entry that represents aggregated performance (e.g., average execution time) of all instances of the virtualization service operation that were completed during an interval (e.g., 1 hour). Aggregated execution metrics may be stored for any number of historical intervals. Since aggregated execution metrics are effectively compressed, diagnostic information about a variety of different types of virtualization service operations may be stored and used to diagnose and repair underperforming components.

Generating, aggregating, and querying virtualization service execution metrics for cloud diagnostics at scale

A computer system that includes at least one host device comprising at least one processor. The at least one processor is configured to implement, in a host operating system (OS) space, a teamed network interface card (NIC) software program that provides a unified interface to host OS space upper layer protocols including at least a remote direct memory access (RDMA) protocol and an Ethernet protocol. The teamed NIC software program provides multiplexing for at least two data pathways. The at least two data pathways include an RDMA data pathway that transmits communications to and from an RDMA interface of a physical NIC, and an Ethernet data pathway that transmits communications to and from an Ethernet interface of the physical NIC through a virtual switch that is implemented in a host user space and a virtual NIC that is implemented in the host OS space.

Userspace networking with remote direct memory access

Examples described herein generally relate to hosting kernel isolated containers within a virtual machine. A server includes a physical processor and a physical computer memory storing executable code, the executable code providing a host virtual machine including a kernel and at least one kernel isolated container within the host virtual machine. The server includes a physical network interface controller, NIC, including a first physical NIC switch and a second physical NIC switch. The first physical NIC switch is configured to distribute incoming data packets to a first plurality of functions including a physical function and virtual functions. At least one of the virtual functions is assigned to the host virtual machine. The second physical NIC switch is configured to distribute the incoming data packets for the host virtual machine to a second plurality of virtual functions including a respective virtual function assigned to an individual kernel isolated container.

Computer device including nested network interface controller switches

Examples described herein generally relate to hosting virtual memory backed kernel isolated containers. A server includes at least one physical processor and at least one physical computer memory addressable via physical memory addresses. The at least one physical computer memory stores executable code configured to provide at least one host including a kernel and at least one kernel isolated container within the at least one host. The host allocates virtual memory having virtual memory addresses to a respective container of the at least one kernel isolated container. The host pins a subset of the virtual memory addresses to a subset of the physical memory addresses. The host performs a direct memory access operation or device memory-mapped input-output operation of the respective container on the subset of the physical memory addresses. At least part of the physical computer memory that is not pinned is oversubscribed.

Pinned physical memory supporting direct memory access for virtual memory backed containers

Embodiments of network processing resource management in computing devices are disclosed therein. In one embodiment, a method includes receiving a request from a network interface controller to perform network processing operations at a first core of a main processor for packets assigned by the network interface controller to a queue of a virtual port of the network interface controller. The method also includes determining whether the first core has a utilization level higher than a threshold when performing the network processing operations to effect processing and transmission of the packets. If the first core has a utilization level higher than the threshold, the method includes issuing a command to the network interface to modify affinitization of the queue from the first core to a second core having a utilization level lower than the threshold.

Management of resource affinities in computing systems having multiple cores

A packet monitoring application instantiated on a server hosting a virtualized network stack is utilized to track data packet propagations and drops at each component within the network stack to reduce the amount of time to identify a root cause for latency issues. The packet monitoring application can be selectively enabled or disabled by an administrator. Components within the virtualized network stack report packet drops and successful packet propagations to the packet monitoring application, which can filter the packets based on input parameters. Thus, a user can select at what level of granularity to filter packets within the virtualized network stack while being able to assess each packet's traversal through each component within the network stack. The packet monitoring application can also perform post-processing of on the filtered data packets to determine latency among components or sections of the virtualized network stack.

Packet drop detection in local networking stack through packet correlation

Embodiments of hybrid network processing load distribution in a computing device are disclosed therein. In one embodiment, a method includes receiving, at a main processor, an indication from the network interface controller to perform network processing operations for first and second packets in a queue of a virtual port of the network interface controller, and in response to receiving the request, assigning multiple cores for performing the network processing operations for the first and second packets, respectively. The method also includes performing the network processing operations at the multiple cores to effect processing and transmission of the first and second packets to first and second applications, respectively, both the first and second applications executing in a virtual machine hosted on the computing device.

Hybrid network processing load distribution in computing systems

A method of communicating data traffic including data packets through a virtual switch on a host device is provided. The data traffic flowing through the virtual switch is monitored. The data traffic includes at least virtual machine data traffic flowing to and from virtual machine processes on the host device and host data traffic flowing to and from host operating system processes on the host device. Each of the data packets are designated as virtual machine data traffic or host data traffic based on an evaluation of the Media Access Controller (MAC) address of each of the one or more data packets of the monitored data traffic. Virtual machine data traffic is directed through a packet processor as the virtual machine data traffic traverses the virtual switch. Host data traffic is directed to bypass the packet processor as the host data traffic traverses the virtual switch.

Virtual switch bypass

Methods and devices for determining settings for a virtual machine may include partitioning a physical network into a plurality of traffic classes. The methods and devices may include determining at least one virtual enhanced transmission selection (ETS) setting for one or more virtual machines, wherein the virtual ETS setting includes at least one virtual traffic class that corresponds to one of the plurality of traffic classes. The methods and devices may include transmitting a notification to the one or more virtual machines identifying the virtual ETS setting.

Virtualizing datacenter bridging settings for virtual network adapters

Embodiments of network processing resource management in computing devices are disclosed therein. An example method includes receiving a request from a network interface controller to perform network processing operations at a first core of a main processor for packets assigned by the network interface controller to a queue of a virtual port of the network interface controller. The method also includes determining whether the first core has a utilization level higher than a threshold when performing the network processing operations to effect processing and transmission of the packets. If the first core has a utilization level higher than the threshold, the method includes issuing a command to the network interface to modify affinitization of the queue from the first core to a second core having a utilization level lower than the threshold.

Modification of queue affinity to cores based on utilization

Embodiments relate to hypervisors that provide hardware isolated virtualization environments (HIVEs) such as containers and virtual machines (VMs). A first HIVE includes a first virtual network interface card (NIC) and a second HIVE includes a second virtual NIC. Both virtual NICs are backed by the same physical NIC. The physical NIC has an Internet Protocol (IP) address. The virtual NICs are assigned the same IP address as the physical NIC. A networking stack of the hypervisor receives inbound packets addressed to the IP address. The networking stack steers the inbound packets to the virtual NICs according to tuples of the inbound packets. Packets emitted by the virtual NICs comprise the IP address, pass through the network stack, and are transmitted by the physical NIC with headers comprising the IP address.

Hypervisor flow steering for address sharing

Techniques of network traffic management in a computing device are disclosed. One example method includes receiving, at a main processor, a request from a network interface controller to perform network processing operations for packets assigned by the network interface controller to a queue of a virtual port of the network interface controller. The method also includes, in response to receiving the request, causing one of multiple cores of the main processor with which the queue of the virtual port is affinitized to perform the network processing operations to effect processing and transmission of the packets to an application executing in a virtual machine hosted on the computing device.

Network traffic management with queues affinitized to one or more cores

Various innovations for servicing of a virtual switch in a virtual networking layer are presented. The innovations include new architectures for a virtual networking layer and new operations performed when servicing a virtual switch. In some example implementations, during virtual switch servicing, interruption to actual network connectivity is minimal&#x2014;below a timeout threshold that signifies failure of a network connection. Connections for a host, VMs, and physical network adapter can be maintained while the virtual switch is serviced. Although some interruption to actual network connectivity happens, apparent connectivity (for a VM) over a connection between the VM and the virtual switch can be maintained during the servicing of the virtual switch. Similarly, apparent connectivity (for a host) over a connection between the host and the virtual switch can be maintained during the servicing of the virtual switch.

Maintaining apparent connectivity during virtual switch servicing

Embodiments of hybrid network processing load distribution in a computing device are disclosed therein. In one embodiment, a method includes receiving, at a main processor, an indication from the network interface controller to perform network processing operations for first and second packets in a queue of a virtual port of the network interface controller, and in response to receiving the request, assigning first and second cores for performing the network processing operations for the first and second packets, respectively. The method also includes performing the network processing operations at the first and second cores to effect processing and transmission of the first and second packets to first and second applications, respectively, both the first and second applications executing in a virtual machine hosted on the computing device.

Growing community of inventors

Bellevue, WA, United States of America

Omar Cardona