Abstract: An example system obtains a request to deploy a cloud template, sorts a plurality of cloud zones in a list based on cost, the plurality of cloud zones being candidates to execute the cloud template, selects a first one of the cloud zones in the list based on the first one of the cloud zones corresponding to a first cost that is lower than second costs of other ones of the cloud zones in the list, and deploys the cloud template on one or more instances of the computing resources corresponding to the first one of the cloud zones.

Abstract: Example methods and systems for health-aware overlay packet forwarding are described. In one example, a first computer system may monitor multiple paths between a first tunnel endpoint (TEP) group and a second TEP group. The first computer system may update health information to indicate that at least a first path is healthy, and a second path is unhealthy. In response to detecting an egress packet that is associated with a packet flow, the first computer system may map the egress packet to the first path instead of the second path based on the health information. The first computer system may generate and forward an encapsulated packet along the first path towards the second computer system, wherein the encapsulated packet includes the egress packet and an outer header that is addressed from a first TEP to a second TEP associated with the first path.

Abstract: A method for network address management is provided. Embodiments include determining a creation of a namespace associated with a cluster of computing devices, wherein a subset of computing resources of the cluster of computing devices is allocated to the namespace. Embodiments include assigning, to the namespace, a network address pool comprising a plurality of network addresses in a subnet, wherein the assigning causes the plurality of network addresses to be reserved exclusively for the namespace. Embodiments include receiving an indication that a pod is added to the namespace. Embodiments include, in response to the receiving of the indication, assigning a network address from the network address pool to the pod.

Abstract: Some embodiments provide a method for operating a physical server in a network. The method stores multiple copies of a virtual machine (VM) image at a network-accessible storage. The method uses a first copy of the VM image as a virtual disk to execute a VM on a hypervisor of a first physical computing device. The method uses a second copy of the VM image as a virtual disk accessible via a smart network interface controller (NIC) of a second physical computing device to execute an operating system of the second physical computing device.

Abstract: A method of collecting health check metrics for a network is provided. The method, at a deep packet inspector on a physical host in a datacenter, receives a copy of a network packet from a load balancer. The packet includes a plurality of layers. Each layer corresponds to a communication protocol in a plurality of communication protocols. The method identifies an application referenced in the packet. The method analyzes the information in one or more layers of the packet to determine metrics for the source application. The method sends the determined metrics to the load balancer.

Abstract: Some embodiments of the invention provide a method of migrating a VM from a first host computer to a second host computer, the first host computer having a first PNIC that performs at least one of network forwarding operations and middlebox service operations for the VM. At an RDMA client executing on a set of one or more processors of the first host computer, the method directs an RDMA server executing on the first PNIC to provide networking state data associated with at least one of network forwarding operations and middlebox service operations that the first PNIC performs for the VM. The provided networking state data resides in a memory of the first PNIC that is accessible to the RDMA server. At the RDMA client, the method provides the obtained networking state data to the second host computer as part of a data migration that is performed to migrate the VM from the first host computer to the second host computer.

Abstract: Some embodiments provide a method of providing distributed storage services to a host computer from a network interface card (NIC) of the host computer. At the NIC, the method accesses a set of one or more external storages operating outside of the host computer through a shared port of the NIC that is not only used to access the set of external storages but also for forwarding packets not related to an external storage. In some embodiments, the method accesses the external storage set by using a network fabric storage driver that employs a network fabric storage protocol to access the external storage set. The method presents the external storage as a local storage of the host computer to a set of programs executing on the host computer. In some embodiments, the method presents the local storage by using a storage emulation layer on the NIC to create a local storage construct that presents the set of external storages as a local storage of the host computer.

Abstract: An example system may include a first endpoint executing a remote collector and a second endpoint in communication with the first endpoint. The remote collector may monitor the second endpoint. The remote collector may include an agent installation unit to install a monitoring agent with configuration data on the second endpoint. The configuration data may specify a configuration for the monitoring agent to monitor a first program executing in the second endpoint. Further, the second endpoint may include a buffer limit configuration unit to execute the monitoring agent in a test mode to determine a first number of metrics to be collected in one cycle based on the configuration data. Furthermore, the buffer limit configuration unit may configure a buffer limit of the monitoring agent based on the first number of metrics and, upon configuring the buffer limit, enable the monitoring agent to monitor the first program.

Abstract: Some embodiments provide a novel method for defining a set of policies for a set of applications executing on a host computer of a software-defined network (SDN). The method configures, on a physical network interface card (PNIC) connected to the host computer, a network adapter to create a logical port that connects an interface of the host computer to a virtual distributed switch (VDS) executing on the PNIC. The method defines the set of policies based on the logical port for the VDS to apply to data message flows sent from the set of applications on the host computer to one or more other host computers of the SDN.

Abstract: Some embodiments provide a novel method for configuring edge routers in a first network. The method configures on a first compute node of the first network (1) a first higher-level edge router and (2) a set of lower-level edge routers. Each lower-level edge router is configured for a different set of subnetworks defined in the first network and is connected to an external second network through the first higher-level edge router. The method detects a condition that requires a particular lower-level edge router for a particular subnetwork to be moved to another compute node. The method configures the particular lower-level edge router to operate on a second compute node below a second higher-level edge router operating on the second compute node to connect the particular lower-level edge router to the external second network.

Abstract: For a network control system that receives, from a user, logical datapath sets that logically express desired forwarding behaviors that are to be implemented by a set of managed switching elements, a controller for managing several managed switching elements that forward data in a network that includes the managed switching elements is described. The controller includes a set of modules for detecting a change in one or more managed switching elements and for updating logical datapath set based on the detected change. The logical datapath set is for subsequent translation into a set of physical forwarding behaviors of the managed switching elements.

Abstract: Container images are managed in a clustered container host system with a shared storage device. Hosts of the system each include a virtualization software layer that supports execution of virtual machines (VMs), one or more of which are pod VMs that have implemented therein a container engine that supports execution of containers within the respective pod VM. A method of deploying containers includes determining, from pod objects published by a master device of the system and accessible by all hosts of the system, that a new pod VM is to be created, creating the new pod VM, and spinning up one or more containers in the new pod VM using images of containers previously spun up in another pod VM, wherein the images of the containers previously spun up in the other pod VM are stored in the storage device.

Abstract: The disclosure provides an approach for dynamic centralized model compilation. Embodiments include receiving, from a client, a request for a machine learning model, wherein the request indicates either one or more attributes comprising one or more of a hardware characteristic, a target precision, or a compiler characteristic, or that one or more default behaviors should be used to compile the machine learning model. Embodiments include determining a compiler for the machine learning model based on the one or more attributes or the one or more default behaviors, wherein the compiler is stored in a registry. Embodiments include compiling the machine learning model using the compiler. Embodiments include providing the compiled machine learning model to the client in response to the request.

Abstract: Disclosed are various embodiments for distributed resource scheduling. An eviction request from a first host is received. The eviction request comprises data regarding a virtual machine to be migrated from the first host. The eviction request is then broadcast to a plurality of hosts. A plurality of responses are received from the plurality of hosts, each response comprising a score representing an ability of a respective one of the plurality of hosts to act as a new host for the virtual machine. A second host is selected from the plurality of hosts to act as the new host for the virtual machine based at least in part on the score in each of the plurality of responses. Then, a response is sent to the first host, the response containing an identifier of the second host.

Abstract: Some embodiments provide a novel method for analyzing data collected by a vehicle at a first location. On a first device operating in the vehicle, the method instantiates a virtual desktop first virtual machine (VM) for remote use of the first device by a remote user that uses a remote second device at a remote second location. The method instantiates a data collection second VM for local use of the first device to collect a set of data at the first location and store the set of data at the vehicle. The method facilitates interaction between the virtual desktop first VM and the remote user over a network for the remote user to access the data collection second VM to analyze the set of data without needing to forward the set of data over the network to the remote second device.

Abstract: Components of a distributed data object are synchronized using streamlined tracking metadata. A target component of the distributed data object is detected as it becomes available and stale. A source component that is up-to-date and that mirrors the address space of the detected target component is identified. A set of mapped address ranges and a set of unmapped address ranges of the identified source component are obtained. A mapped address range of the target component that corresponds with an unmapped address range of the source component is identified. The identified mapped address range of the target component is then synchronized with the corresponding unmapped address range of the source component. Thus, unmapped address ranges are synchronized without using tracking metadata of the source component.

Abstract: Disclosed embodiments pertain to support input/output modules for container volumes. An input/output (I/O) request can be received from a containerized application. A container volume targeted by the I/O request can be identified. A determination is then made that the container volume is associated with one or more I/O modules based on a stored mapping of container volumes to I/O modules. Data associated with the I/O request is sent to the one or more I/O modules for processing. Processed data can be received from the one or more I/O modules, and the I/O request is fulfilled using the processed data. In certain embodiments, a write I/O request is fulfilled by writing the processed data to a virtual disk file for the container volume, and a read I/O request is fulfilled by returning the original data after reversing the processing to the containerized application.

Abstract: The disclosure provides an approach for a hardware agnostic paravirtualized machine learning (ML) and artificial intelligence (AI) accelerator in a virtual environment. A method includes running a ML application on a first virtual computing instance (VCI) on a host; generating a set of tensors based on one or more computations associated with the ML application; generating a computational graph for the ML application based on the set of tensors; providing, by a paravirtualized ML accelerator running on the host, the computational graph to a driver VCI running on the host; and executing the one or more computations associated with the ML application using at least one of the one or more physical hardware ML accelerators associated with the paravirtualized ML accelerator.

Abstract: Some embodiments provide a novel method for performing services on a host computer that executes several data compute nodes (DCNs). The method receives, at a module executing on the host, a data message associated with a DCN executing on the host. The method supplies the data message to a service virtual machine (SVM) that executes on the host and on which several service containers execute. One or more of the service containers then perform a set of one or more services on the data message. The method then receives an indication from the SVM that the set of services has been performed on the data message.

Abstract: Policies are defined for a container cluster that is configured by a first software defined network (SDN) controller cluster. A second SDN controller cluster for defining service policies that are not defined by the first SDN controller cluster receives, from a set of one or more adapters deployed in the container cluster for the second SDN controller cluster, resource identifiers for several resources of the container cluster. The second SDN controller cluster uses the resource identifiers to define a set of service policies. Then, the second SDN controller cluster distributes the set of service policies to a set of network elements to enforce the set of service policies on data messages associated with machines deployed in the container cluster configured by the first SDN controller cluster.