Abstract: Techniques are described for a virtual smart network interface card to facilitate data transmission in an edge device providing cloud-computing operations. A private virtual network data plane hosted within an execution environment of the cloud-computing edge device can receive a request to create a virtual network interface. The request can include a networking command. The private virtual network data plane can send the networking command to a command proxy module. The command proxy module can then use the networking command to generate the virtual network interface within a first networking namespace of a host operating system executing on the cloud-computing edge device.

Abstract: Techniques are disclosed for migrating one or more services from an edge device to a cloud computing environment. In one example, a migration service receives a request to migrate a first set of services from the edge device to the cloud computing environment. The migration service identifies a hardware profile of a computing device (or devices) of the cloud computing environment that matches the edge device, and then configures the computing device to execute a second set of services that corresponds to the first set of services. The migration service establishes a communication channel between the edge device and the computing device, and then executes a set of migration operations such that the second set of services is configured to execute as the first set of services. The computing device may operate in a virtual bootstrap environment or dedicated region of the cloud computing environment.

Abstract: A method may include transmitting a request for metadata associated with a compute instance and receiving, by a computing system, metadata associated with the compute instance signed with a private key. The private key may be associated with a public key. The method may include receiving a request to access a cloud resource and transmitting the request for the metadata. The method may also include receiving the metadata. The metadata may indicate that the compute instance is hosted on the computing system. The method may also include transmitting, to an instance principal service, a request for an instance principal certificate. The request may include the metadata signed with the private key and be cryptographically verified by the instance principal service using the public key. The method may also include receiving the instance principal certificate and providing access to the could resource based on the instance principal certificate.

Abstract: Techniques and apparatus for data networking are described. In one example, a method of queuing Remote Direct Memory Access (RDMA) packets includes receiving a first RDMA packet having a first quality-of-service (QOS) data field; based on a value of the first QoS data field, queueing the first RDMA packet in a first queue of a plurality of queues; receiving a second RDMA packet having a second QoS data field; and based on a value of the second QoS data field, queueing the second RDMA packet in a second queue of the plurality of the queues, the second queue being different than the first queue.

Abstract: Techniques are described for implementing a virtual smart network interface card to facilitate data transmission in an edge device providing cloud-computing operations. An edge device can implement a private virtual network that includes a private virtual network data plane. The edge device can execute a virtual machine to be connected to the private virtual network. To establish the connection, the edge device can generate a virtual network interface that includes a first endpoint and a second endpoint and is hosted within the private virtual network data plane. The edge device can associate the first endpoint with the virtual machine and associate the second endpoint with an orchestration module of the private virtual network data plane. The virtual machine can then send a data packet to the orchestration module via the virtual network interface.

Abstract: Techniques and apparatus for data networking are described. In one example, a method includes receiving a first Layer-2 Remote Direct Memory Access (RDMA) packet which includes a virtual local area network (VLAN) tag and a quality-of-service (QoS) data field; converting the first Layer-2 RDMA packet to a first Layer-3 encapsulated packet; and forwarding the first Layer-3 encapsulated packet to a switch fabric. In this method, the converting includes adding at least one header to the first Layer-2 RDMA packet, where the at least one header includes: a virtual network identifier that is based on information from the VLAN tag, and a QoS value that is based on information from the QoS data field.

Abstract: Techniques and apparatus for data networking are described. In one example, a method of queuing Remote Direct Memory Access (RDMA) packets includes receiving a first RDMA packet having a first quality-of-service (QoS) data field; based on a value of the first QoS data field, queueing the first RDMA packet in a first queue of a plurality of queues; receiving a second RDMA packet having a second QoS data field; and based on a value of the second QoS data field, queueing the second RDMA packet in a second queue of the plurality of the queues, the second queue being different than the first queue.

Abstract: Techniques and apparatus for data networking are described. In one example, a method includes receiving a first Layer-2 Remote Direct Memory Access (RDMA) packet which includes a virtual local area network (VLAN) tag and a quality-of-service (QoS) data field; converting the first Layer-2 RDMA packet to a first Layer-3 encapsulated packet; and forwarding the first Layer-3 encapsulated packet to a switch fabric. In this method, the converting includes adding at least one header to the first Layer-2 RDMA packet, where the at least one header includes: a virtual network identifier that is based on information from the VLAN tag, and a QoS value that is based on information from the QoS data field.

Abstract: Described herein is a network fabric including a plurality of graphical processing unit (GPU) clusters that are communicatively coupled with one another via a plurality of switches arranged in a hierarchical structure including a first tier of switches, a second tier of switches, and a third tier of switches. One or more switches are selected from the third tier of switches to form a set of target switches, where each target switch receives address information of each GPU included in the plurality of GPU clusters. Each target switch generates, a plurality of sets of address information by filtering received address information based on a condition and transmits the plurality of sets of address information to each switch included in the first tier of switches, wherein the switch stores a subset of the plurality of sets of address information in accordance with the condition.

Abstract: Each host machine of a plurality of host machines stores hierarchical locality information for the host machine that identifies at least a rack comprising the host machine, and a block of a plurality of blocks hosting the rack. The host machine is associated with one or more graphical processing units (GPUs), and wherein GPUs included in a first block operate at a first speed and GPUs included in a second block operate at a second speed that is different than the first speed. Responsive to receiving a request requesting execution of a workload, one or more host machines are identified as being available for executing the workload, and the hierarchical locality information and linkage information of the one or more host machines is provided in response to the request.

Abstract: A plurality of GPU clusters are communicatively coupled with one another via a plurality of network devices arranged in a hierarchical structure, wherein the GPU clusters includes at least a first GPU cluster operating at a first speed and a second GPU cluster operating at a second speed that is different than the first speed. A routing policy is configured for each network device, wherein the configuring includes establishing a mapping of each incoming port-link of the network device to a unique outgoing port-link of the network device. For a packet transmitted by a GPU of a host machine and received by a first network device, an incoming port-link of the first network device is determined on which the packet was received and based on the configuring, an outgoing port-link is identified that corresponds to the incoming port-link. The packet is forwarded on the outgoing port-link of the network device.

Abstract: Described herein is a network fabric including a plurality of graphical processing unit (GPU) clusters. The plurality of GPU clusters includes at least a first GPU cluster operating at a first speed and a second GPU cluster operating at a second speed that is different than the first speed. The network fabric includes a plurality of blocks, wherein each block includes: (a) one or more racks that host a GPU cluster, and (b) a plurality of switches arranged in a hierarchical structure that communicatively couple the block to other blocks included in the network fabric. Responsive to receiving a request to execute a workload, allocating one or more GPUs from the plurality of GPU clusters to execute the workload.

Abstract: Discussed herein is a framework that provisions for customized processing for different classes of traffic. A network device in a communication path between a source host machine and a destination host machine extracts a tag from a packet received by the network device. The packet originates at a source executing on the source host machine and whose destination is the destination host machine. The tag set by the source and indicative of a first traffic class to be associated with the packet, the first traffic class being selected by the source from a plurality of traffic classes. The network device determines, based on the tag, that the first traffic class corresponds to a latency sensitive traffic and processes the packet using one or more settings configured at the network device for processing packets associated with the first traffic class.

Abstract: Techniques discussed herein relate to providing in-memory workflow management at an edge device (e.g., a computing device distinct from and operating remotely with respect to a data center). The edge device can operate as a computing node in a computing cluster of edge devices and implement a hosting environment (e.g., a distributed data plane). A work request can be obtained by an in-memory workflow manager of the edge device. The work request may include an intended state of a data plane resource (e.g., a computing cluster, a virtual machine, etc.). The in-memory workflow manager can determine the work request has not commenced and initialize an in-memory execution thread to execute orchestration tasks to configure a data plane of the computing cluster according to the intended state. Current state data corresponding to the configured data plane may be provided to the user device and eventually displayed.

Abstract: Techniques discussed herein relate to updating an edge device (e.g., a computing device distinct from and operating remotely with respect to a data center). The edge device can execute a first operating system (OS). A manifest specifying files of a second OS to be provisioned to the edge device may be obtained. The manifest may further specify a set of services to be provisioned at the edge device. One or more data files corresponding to a difference between a first set of data files associated with the first OS and a second set of data files associated with the second OS may be identified. A snapshot of the first OS may be generated, modified, and stored in memory of the edge device to configure the edge device with the second OS. The booting order of the edge device may be modified to boot utilizing the second OS.

Abstract: Discussed herein is a framework that provisions for customized processing for different classes of traffic. A network device in a communication path between a source host machine and 5 a destination host machine extracts a tag from a packet received by the network device. The packet originates at a source executing on the source host machine and whose destination is the destination host machine. The tag set by the source and indicative of a first traffic class to be associated with the packet, the first traffic class being selected by the source from a plurality of traffic classes. The network device determines the first traffic class based on the tag extracted from the packet and 10 processes the packet based on the first traffic class.

Abstract: Discussed herein is a framework that provisions for customized processing for different classes of traffic. A network device in a communication path between a source host machine and a destination host machine extracts a tag from a packet received by the network device. The packet originates at a source executing on the source host machine and whose destination is the destination host machine. The tag set by the source and indicative of a first traffic class to be associated with the packet, the first traffic class being selected by the source from a plurality of traffic classes. The network device determines, based on the tag, that the first traffic class corresponds to a bandwidth sensitive traffic and processes the packet using one or more settings configured at the network device for processing packets associated with the first traffic class.

Abstract: Techniques discussed herein relate to providing in-memory workflow management at an edge device (e.g., a computing device distinct from and operating remotely with respect to a data center). The edge device can operate as a computing node in a computing cluster of edge devices and implement a hosting environment (e.g., a distributed data plane). A work request can be obtained by an in-memory workflow manager of the edge device. The work request may include an intended state of a data plane resource (e.g., a computing cluster, a virtual machine, etc.). The in-memory workflow manager can determine the work request has not commenced and initialize an in-memory execution thread to execute orchestration tasks to configure a data plane of the computing cluster according to the intended state. Current state data corresponding to the configured data plane may be provided to the user device and eventually displayed.

Abstract: Discussed herein are techniques that utilize locality information of host machines included in a cluster network for the execution of graphical processing unit based workloads. For each host machine of a plurality of host machines, locality information for the host machine is stored therein. The locality information for a host machine identifies a rack comprising the host machine. Responsive to receiving a request requesting execution of a workload, one or more host machines of the plurality of host machines are identified as being available for executing the workload. For each of the one or more host machines, the locality information for the host machine is obtained. Further, linkage information of the one or more host machines is identified. The locality information and the linkage information of the one or more host machines is provided in response to the request.

Abstract: Discussed herein are techniques that utilize hierarchical locality information of host machines included in a cluster network for the execution of general workloads. Hierarchical locality information for each host machine of a plurality of host machines is stored. The hierarchical locality information for a host machine identifying, for each locality of a plurality of localities, location information for the locality. Responsive to receiving a request requesting execution of a workload, the hierarchical locality information for the plurality of host machines is obtained and provided (e.g., to a customer) in response to the request.