Cisco Technology, Inc.

Raghunath Nambiar

Karthik Kulkarni

Debojyoti Dutta

Madhav V Marathe

Yathiraj B Udupi

Senhua Huang

Rajesh Shroff

Sean Mckeown

Disclosed are systems, methods, and computer-readable storage media for infrastructure aware adaptive resource allocation. A resource manager can receive a request to allocate a first container instance in a network. The resource manager can determine a set of candidate computing nodes in the network that are capable of hosting the first container instance and request, from an infrastructure monitor that received infrastructure data from the network, health metrics for the set of candidate computing nodes. The resource manager can select, based on the health metrics for the set of candidate computing nodes, an optimal computing node to host the first container instance, and allocate the first container instance on the optimal computing node.

Infrastructure aware adaptive resource allocation

A system and a method for prioritizing network traffic using application awareness and network awareness in a network environment is disclosed. An exemplary method can include receiving a network traffic priority for network traffic associated with a job performed by a distributed application; and based on the network traffic priority, assigning the network traffic to a network classification model, such as an access control list, associated with a network element for directing the network traffic in a network. The network classification model is associated with a network traffic priority class having defined quality of service (QoS) features, such that the network traffic traverses the network according to the defined QoS features.

Distributed application framework for prioritizing network traffic using application priority awareness

Systems and methods are described for allocating resources in a cloud computing environment. The method includes receiving a computing request, the request for use of at least one virtual machine and a portion of memory. In response to the request, a plurality of hosts is identified and a cost function is formulated using at least a portion of those hosts. Based on the cost function, at least one host that is capable of hosting the virtual machine and memory is selected.

Optimizing placement of virtual machines

The subject technology provides a drillable time-series heat map, which combines information of separate network element (e.g., switch, router, server or storage) and relates them together through impact zones to correlate network wide events and the potential impact on the other units in the network. The subject technology also brings together the network and its components, the distributed application(s) and a heat map controller to proactively communicate with one another to disseminate information such as failures, timeouts, new jobs, etc. For an individual job (e.g., for a distributed application), the subject technology may monitor consumption of resources during different phases of execution to provide individual job profile data that could be presented as a drillable heat map. The heat map, in this regard, nay include resource utilization heat metrics of resources such as CPU, Input/Output (I/O), memory, etc., in the heat map or graphs and presented along with network activity.

Annotation of network activity through different phases of execution

A distributed application framework, along with related systems and/or methods described herein, can intelligently place data using network knowledge. An exemplary method can include receiving data placement information from a distributed application that identifies a source node of data in a network and a list of potential destination nodes in the network for the distributed application to place the data; for each potential destination node, determining a network latency associated with transferring the data from the source node to the potential destination node using network metrics associated with the network; and sending the determined network latencies to the distributed application, such that the distributed application can assign the data to one of the potential destination nodes based on the determined network latencies.

Distributed application framework that uses network and application awareness for placing data

Systems, methods, and computer program products to perform an operation comprising receiving a priority of a distributed computing job, an intermediate traffic type of the distributed computing job, and a set of candidate compute nodes available to process the distributed computing job, the candidate compute nodes each available to process at least one input split of the distributed computing job, and selecting a mapper node from the candidate compute nodes, for one of the input splits, wherein the mapper node is selected based on the priority and the intermediate traffic type of the distributed computing job, wherein the mapper compute node is further selected upon determining that the mapper node is not affected by an error, and a resource utilization score for the mapper node does not exceed a utilization threshold.

Adaptive datacenter topology for distributed frameworks job control through network awareness

In one embodiment, data indicative of the size of an intermediate data set generated by a first resource device is received at a computing device. The intermediate data set is associated with a virtual machine to process the intermediate data set. A virtual machine configuration is determined based on the size of the intermediate data set. A second resource device is selected to execute the virtual machine based on the virtual machine configuration and on an available bandwidth between the first and second resource devices. The virtual machine is then assigned to the second resource device to process the intermediate data set.

Allocating resources for multi-phase, distributed computing jobs

In one embodiment, a device receives information regarding a data set to be processed by a map-reduce process. The device generates a set of virtual clusters for the map-reduce process based on network bandwidths between nodes of the virtual clusters, each node of the virtual cluster corresponding to a resource device, and associates the data set with a map-reduce process task. The device then schedules the execution of the task by a node of the virtual clusters based on the network bandwidth between the node and a source node on which the data set resides.

Task scheduling using virtual clusters

The present disclosure relates to assignment or generation of reducer virtual machines after the 'map' phase is substantially complete in MapReduce. Instead of a priori placement, distribution of keys after the 'map' phase over the mapper virtual machines can be used to efficiently reducer tasks in virtualized cloud infrastructure like OpenStack. By solving a constraint optimization problem, reducer VMs can be optimally assigned to process keys subject to certain constraints. In particular, the present disclosure describes a special variable matrix. Furthermore, the present disclosure describes several possible cost matrices for representing the costs determined based on the key distribution over the mapper VMs (and other suitable factors).

Growing community of inventors

San Ramon, CA, United States of America

Raghunath Nambiar