Microsoft Technology Licensing, LLC

Karthikeyan Subramanian

Shandan Zhou

Thomas Moscibroda

Si Qin

Saurabh Agarwal

Gowtham Natarajan

Yong Xu

Qingwei Lin

Gurpreet Virdi

Yunus Mohammed

Sorin Iftimie

Ajay Mani

Hang Dong

Bo Qiao

Nisarg Tarakkumar Sheth

Sanjay Ramanujan

Marko Lalic

Wanghai Gu

Sandeep Ramji

Paul Naveen Selvaraj

Yuwen Yang

Asad Yaqoob

Todd F Pfleiger

John Lawrence Miller

Gaurav Jagtiani

Murtaza Muidul Huda Chowdhury

Eli Cortez

Murali Mohan Chintalapati

Ambrose Thomas Treacy

Zainab Hakim

Abhay Sudhir Ketkar

Shanti Kemburu

Luke Rafael Rodriguez

Abhisek Pan

Shweta Balkrishna Patil

Hemant Malik

Yancheng Chen

Dalianna Victoria Vaysman

Sam Prakash Bheri

Fernando Gonzalez Todisco

Chunxiao Jia

Vi Lam Nguyen

Rostislav Sudakov

Xing Wen Wang

Christopher Cowdery

Yash Purohit

Neha Keshari

Vinayak Ramnath Karnataki

Maitreyee Ramprasad Joshi

Shuishi Yang

Michal Jama

Valentina Li

Vipin Kumar

Koon Hui Geoffrey Goh

Karel Trueba Nobregas

The present disclosure relates to systems, methods, and computer readable media for predicting surplus capacity on a set of server nodes and determining a quantity of deferrable virtual machines (VMs) that may be scheduled over an upcoming period of time. This determination of VM quantity may be determined while minimizing risks associated with allocation failures on the set of server nodes. This disclosure described systems that facilitate features and functionality related to improving utilization of surplus resource capacity on a plurality of server nodes by implementing VMs having some flexibility in timing of deployment while also avoiding significant risk caused as a result of over-allocated storage and computing resources. In one or more embodiments, the quantity of deferrable VMs is determined and scheduled in accordance with rules of a scheduling policy.

Allocating computing resources for deferrable virtual machines

The present disclosure relates to systems, methods, and computer readable media for predicting expansion failures and implementing defragmentation instructions based on the predicted expansion failures and other signals. For example, systems disclosed herein may apply a failure prediction model to determine an expansion failure prediction associated with an estimated likelihood that deployment failures will occur on a node cluster. The systems disclosed herein may further generate defragmentation instructions indicating a severity level that a defragmentation engine may execute on a cluster level to prevent expansion failures while minimizing negative customer impacts. By uniquely generating defragmentation instructions for each node cluster, a cloud computing system can minimize expansion failures, increase resource capacity, reduce costs, and provide access to reliable services to customers.

Predicting expansion failures and defragmenting cluster resources

A computer implemented method includes receiving telemetry data corresponding to capacity health of nodes in a cloud based computing system. The received telemetry data is processed via a prediction engine to provide predictions of capacity health at multiple dimensions of the cloud based computing system. Node recoverability information is received and node recovery execution is initiated as a function of the representations of capacity health and node recoverability information.

Capacity aware cloud environment node recovery system

A cloud computing capacity management system can include a fine-grained admission control layer, a policy engine, and an enforcement layer. The fine-grained admission control layer can be configured to ingest capacity signals and create a capacity mitigation policy, based at least in part on the capacity signals, to protect available capacity of a cloud computing system for prioritized users. The capacity mitigation policy can be directed to users of the cloud computing system. The policy engine can be configured to control how the capacity mitigation policy is applied to the cloud computing system. The enforcement layer can be configured to handle incoming resource requests and to enforce resource limits based on the capacity mitigation policy as applied by the policy engine.

Cloud computing capacity management system using automated fine-grained admission control

The present disclosure relates to systems, methods, and computer readable media for predicting deployment growth on one or more node clusters and selectively permitting deployment requests on a per cluster basis. For example, systems disclosed herein may apply tenant growth prediction system trained to output a deployment growth classification indicative of a predicted growth of deployments on a node cluster. The system disclosed herein may further utilize the deployment growth classification to determine whether a deployment request may be permitted while maintaining a sufficiently sized capacity buffer to avoid deployment failures for existing deployments previously implemented on the node cluster. By selectively permitting or denying deployments based on a variety of factors, the systems described herein can more efficiently utilize cluster resources on a per-cluster basis without causing a significant increase in deployment failures for existing customers.

Allocating cloud resources in accordance with predicted deployment growth

A method for minimizing allocation failures in a cloud computing system without overprovisioning may include determining a predicted supply for a virtual machine series in a system unit of the cloud computing system during an upcoming time period. The predicted supply may be based on a shared available current capacity and a shared available future added capacity for the virtual machine series in the system unit. The method may also include predicting an available capacity for the virtual machine series in the system unit during the upcoming time period. The predicted available capacity may be based at least in part on a predicted demand for the virtual machine series in the system unit during the upcoming time period and the predicted supply. The method may also include taking at least one mitigation action in response to determining that the predicted demand exceeds the predicted supply during the upcoming time period.

Capacity management in a cloud computing system using virtual machine series modeling

A method for evaluating at least one potential policy for an IaaS system may include determining a predicted workload for the IaaS system based on at least one generative model corresponding to the IaaS system. The at least one potential policy for the IaaS system may be simulated based on the predicted workload, thereby producing one or more simulation metrics that indicate effects of the at least one potential policy. The performance of the IaaS system may be optimized based on the one or more simulation metrics.

Using a generative model to facilitate simulation of potential policies for an infrastructure as a service system

Embodiments relate to determining whether to take a resource distribution unit (RDU) of a datacenter offline when the RDU becomes faulty. RDUs in a cloud or datacenter supply a resource such as power, network connectivity, and the like to respective sets of hosts that provide computing resources to tenant units such as virtual machines (VMs). When an RDU becomes faulty some of the hosts that it supplies may continue to function and others may become unavailable for various reasons. This can make a decision of whether to take the RDU offline for repair difficult, since in some situations countervailing requirements of the datacenter may be at odds. To decide whether to take an RDU offline, the potential impact on availability of tenant VMs, unused capacity of the datacenter, a number or ratio of unavailable hosts on the RDU, and other factors may be considered to make a balanced decision.

Opportunistic offlining for faulty devices in datacenters

Aspects extend to methods, systems, and computer program products for clustering streamed or batch data. Aspects of the invention include dynamic clustering and labeling of streamed data and/or batch data, including failures and error logs (user, platform, etc.), latency logs, warning logs, information logs, Virtual Machine (VM) creation data logs, template logs, etc., for use in analysis (e.g., error log analysis). A clustering system can learn from previously identified patterns and use that information to group newer information dynamically as it gets generated. The clustering system can leverage streamed data and/or batch data domain knowledge for preprocessing. In one aspect, a clustering system uses a similarity measure. Based on (e.g., users' configuration of) a similarity threshold, the cluster system (e.g., automatically) assigns/clusters streamed data and/or batch data into groups.

Clustering and labeling streamed data

Techniques for automated capacity managed in distributed computing systems are disclosed herein. In one embodiment, a method includes receiving predicting one or more future usage levels of a computing resource in the distributed computing system based on received data representing historical usage levels of the computing resource and determining whether a currently available capacity of the computing resource in the distributed computing system is depleted beyond a threshold time period based on the one or more future usage levels. In response to determining that the currently available capacity of the computing resource in the distributed computing system is depleted before the threshold time period, the method includes immediately rebooting, reimaging, or performing other recovery actions on one or more out-for-repair hosts that provide the computing resource.

Automated capacity management in distributed computing systems

A distributed computing system. The system separates management and worker functionality into different clusters thus relieving the worker clusters of management overhead. In particular, the system includes a management cluster. The management cluster is configured to perform service deployment of worker nodes, service maintenance of worker nodes, service upgrade and roll-back of worker nodes, service monitoring of worker nodes, and service reporting of worker nodes. The system further includes one or more worker clusters coupled the management cluster, but separate from the management cluster. Each of the worker clusters includes one or more worker nodes that are configured to provide one or more of processing power, storage, and network connectivity but do not exercise one or more of the management configured functions.

Building large scale infrastructure using hybrid clusters

Growing community of inventors

Redmond, WA, United States of America

Karthikeyan Subramanian