Abstract: In some embodiments, a method receives a set of packets for a flow and determines a set of features for the flow from the set of packets. A classification of an elephant flow or a mice flow is selected based on the set of features. The classification is selected before assigning the flow to a network resource in a plurality of network resources. The method assigns the flow to a network resource in the plurality of network resources based on the classification for the flow and a set of classifications for flows currently assigned to the plurality of network resources. Then, the method sends the set of packets for the flow using the assigned network resource.

Abstract: Disclosed are various embodiments for rate proportional scheduling to reduce packet loss in virtualized network function chains. A congestion monitor executed by a first virtual machine executed by a host computing device can detect congestion in a receive queue associated with a first virtualized network function implemented by a first virtual machine. The congestion monitor can send a pause signal to a rate controller executed by a second virtual machine executed by the host computing device. The rate controller can receive the pause signal. In response, the rate controller can pause the processing of packets by a second virtualized network function implemented by the second virtual machine to reduce congestion in the receive queue of the first virtualized network function.

Abstract: Solutions for discovering and onboarding edge devices at scale include: receiving, by a device aggregator, edge device state information including state information for a first edge device; based on at least the state information for the first edge device, configuring the first edge device to perform as a software-defined wide area network (SD-WAN) node; based on at least the edge device state information, determining a first device profile for the first edge device; and transmitting the first device profile to a workload manager. In some examples, the edge device state information includes state information for a second edge device; the second edge device is configured to perform as an SD-WAN node; and a second device profile is determined for the second edge device and transmitted to the workload manager. The workload manager allocates an SD-WAN workload among the first edge device and the second edge device.

Abstract: An example computer system includes a hardware platform including a processing unit and software executing on the hardware platform. The software includes a workload and a scheduler, the workload including a network function chain having network functions, the scheduler configured to schedule the network functions for execution on the processing unit. A downstream network function includes a congestion monitor configured to monitor a first receive queue supplying packets to the downstream network function, the congestion monitor configured to compare occupancy of the first receive queue against a queue threshold. An upstream network function including a rate controller configured to receive a notification from the congestion monitor generated in response to the occupancy of the first receive queue exceeding the queue threshold, the rate controller configured to modify a rate of packet flow between a second receive queue and the upstream network function in response to the notification.

Abstract: Disclosed are various embodiments for optimizing the migration of pages of memory servers in cluster memory systems. To begin, a computing device can mark in a page table of the computing device that a page stored on a first memory host is not present. Then, the computing device can flush a translation lookaside buffer of the computing device. Next, the computing device can copy the page from the first memory host to a second memory host. Moving on, the computing device can update a page mapping table to reflect that the page is stored in the second memory host. Then, the computing device can mark in the page table of the computing device that the page stored in the second memory host is present. Subsequently, the computing device can discard the page stored on the first memory host.

Abstract: In some embodiments, a method receives a set of packets for a flow and determines a set of features for the flow from the set of packets. A classification of an elephant flow or a mice flow is selected based on the set of features. The classification is selected before assigning the flow to a network resource in a plurality of network resources. The method assigns the flow to a network resource in the plurality of network resources based on the classification for the flow and a set of classifications for flows currently assigned to the plurality of network resources. Then, the method sends the set of packets for the flow using the assigned network resource.

Abstract: A framework that may be implemented by a workload orchestration platform for scheduling accelerator-enabled workloads on the accelerators in a cluster is provided. In one set of embodiments, the framework enables the platform to schedule accelerator-enabled workloads based on a multitude of user-provided, fine-grained accelerator requirements. In another set of embodiments, the framework enables the platform to automatically recommend an initial set of accelerator resource requirements for an accelerator-enabled workload and automatically right-size such requirements based on telemetry data collected during the workload's runtime.

Abstract: Some embodiments provide a method of implementing capacity-aware load balancing across a set of data compute nodes (DCNs) by reducing latency for the set of DCNs. From the set of DCNs, the method identifies (1) a first subset of DCNs including DCNs that have a latency that is higher than an average latency computed for the set of DCNs and (2) a second subset of DCNs including DCNs that have a latency that is lower than the average latency computed for the set of DCNs. For each DCN in the first subset of DCNs, the method assigns to the DCN a weight value that corresponds to a target latency computed for the set of DCNs. Based on the assigned weight values for the first subset of DCNs, the method computes an excess weight value to be redistributed across the second subset of DCNs. The method redistributes the computed excess weight value across the second subset of DCNs.

Abstract: Solutions for discovering and onboarding edge devices at scale include: receiving, by a device aggregator, edge device state information including state information for a first edge device; based on at least the state information for the first edge device, configuring the first edge device to perform as a software-defined wide area network (SD-WAN) node; based on at least the edge device state information, determining a first device profile for the first edge device; and transmitting the first device profile to a workload manager. In some examples, the edge device state information includes state information for a second edge device; the second edge device is configured to perform as an SD-WAN node; and a second device profile is determined for the second edge device and transmitted to the workload manager. The workload manager allocates an SD-WAN workload among the first edge device and the second edge device.

Abstract: Disclosed are aspects of network function placement in virtual graphics processing unit (vGPU)-enabled environments. In one example a network function request is associated with a network function. A scheduler selects a vGPU-enabled GPU to handle the network function request. The vGPU-enabled GPU is selected in consideration of a network function memory requirement or a network function IO requirement. The network function request is processed using an instance of the network function within a virtual machine that is executed using the selected vGPU-enabled GPU.

Abstract: Disclosed are various embodiments for improving resiliency and performance of clustered memory. A computing device can acquire a chunk of byte-addressable memory from a cluster memory host. The computing device can then identify an active set of allocated memory pages and an inactive set of allocated memory pages for a process executing on the computing device. Next, the computing device can store the active set of allocated memory pages for the process in the memory of the computing device. Finally, the computing device can store the inactive set of allocated memory pages for the process in the chunk of byte-addressable memory of the cluster memory host.

Abstract: Disclosed are various embodiments for improving the resiliency and performance for clustered memory. A computing device can mark a page of the memory as being reclaimed. The computing device can then set the page of the memory as read-only. Next, the computing device can submit a write request for the contents of the page to individual ones of a plurality of memory hosts. Subsequently, the computing device can receive individual confirmations of a successful write of the page from the individual ones of the plurality of memory hosts. Then, the computing device can mark the page as free in response to receipt of the individual confirmations of the successful write from the individual ones of the plurality of memory hosts.

Abstract: Computer-implemented methods, media, and systems for inter-cluster automated failover and migration of containerized workloads across edges devices are disclosed. One example method includes monitoring telemetry data received from a first software defined wide area network (SD-WAN) edge device that has a workload scheduled, where the telemetry data includes at least one of a health status of the workload or multiple runtime context elements at the first SD-WAN edge device. It is determined that a failure associated with either the first SD-WAN edge device or the workload occurs. A mode of the failure is determined. A remediation process based on the determined mode of the failure and a current state of the workload is performed.

Abstract: Systems and methods are described for providing a graphical user interface (“GUI”) for migrating workloads in a system. The GUI can display the locations of edge devices in the system and workloads running on the edge devices. A user can drag a workload from one edge device to another in the GUI, and in response the system can schedule the workload to be migrated accordingly. Before the migration is performed, the GUI can calculate a change in computing resource usage at both edge devices. The GUI can display the usage data and prompt the user to confirm the migration. If the user confirms, the workload can be deployed at the target edge device and removed from the source edge device.

Abstract: Disclosed are various embodiments for improving the resiliency and performance of clustered memory. A computing device can generate at least one parity page from at least a first local page and a second local page. The computing device can then submit a first write request for the first local page to a first one of a plurality of memory hosts. The computing device can also submit a second write request for the second local page to a second one of the plurality of memory hosts. Additionally, the computing device can submit a third write request for the parity page to a third one of the plurality of memory hosts.

Abstract: Disclosed are various embodiments for improving the resiliency and performance of clustered memory. A computing device can generate at least one parity page from at least a first local page and a second local page. The computing device can then submit a first write request for the first local page to a first one of a plurality of memory hosts. The computing device can also submit a second write request for the second local page to a second one of the plurality of memory hosts. Additionally, the computing device can submit a third write request for the parity page to a third one of the plurality of memory hosts.

Abstract: Some embodiments provide a method of implementing context-aware routing for a software-defined wide-area network, at an SD-WAN edge forwarding element (FE) located at a branch network connected to the SD-WAN. The method receives, from an SD-WAN controller, geolocation route weights for each of multiple cloud datacenters across which a set of application resources is distributed. The application resources are all reachable at a same virtual network address. For each of the cloud datacenters, the method installs a route for the virtual network address between the branch network and the cloud datacenter. The routes have different total costs based at least in part on the geolocation metrics received from the SD-WAN controller. The SD-WAN edge FE selects between the routes to establish connections to the set of application resources.

Abstract: Described herein are systems, methods, and software to manage power consumption in a software build environment. In one implementation, a monitoring service monitors power consumption information associated with a build environment for one or more software components. The monitoring service further identifies one or more trends associated with the power consumption information based at least on the power consumption information satisfying one or more criteria and generates a summary for display that indicates at least the one or more trends. The monitoring service may also identify and display as part of the summary one or more suggestions to improve power consumption based on the one or more trends.

Abstract: Computer-implemented methods, media, and systems for context-sensitive defragmentation and aggregation of containerized workloads running on edge devices are disclosed. One example method includes monitoring telemetry data from multiple software defined wide area network (SD-WAN) edge devices that run multiple workloads, where the telemetry data includes at least one of resource utilization at the multiple SD-WAN edge devices, inter-workload trigger dependency, or inter-workload data dependency among the multiple workloads. It is determined, based on the telemetry data, that at least two of the multiple workloads running on at least two SD-WAN edge devices have the inter-workload trigger dependency or the inter-workload data dependency.

Abstract: Computer-implemented methods, media, and systems for remediation of containerized workloads based on context breach at edge devices are disclosed. One example computer-implemented method includes monitoring telemetry data from a first software defined wide area network (SD-WAN) edge device, where the telemetry data includes multiple context elements at the first SD-WAN edge device. It is determined that a context change occurs for at least one of the context elements at the first SD-WAN edge device. It is determined that due to the context change, the first SD-WAN edge device does not satisfy one or more requirements for running one or more workloads scheduled to run. In response to the determination that the first SD-WAN edge device does not satisfy the one or more requirements, the at least one of the one or more workloads is offloaded from the first SD-WAN edge device to a second SD-WAN edge device.