International Business Machines Corporation

Myung Mun Bae

Robert K Gardner

Ramendra K Sahoo

Jifang Zhang

Chulho Kim

Felipe Knop

Rajeev Sivaram

Paul D DiNicola

Perinkulam I Ganesh

Peter Richard Badovinatz

Jeffrey S Lucash

Esdras E Cruz-Aguilar

Michael Anthony Schmidt

Dimpu K Nath

Savitri Hunasheekatti

Johannes Sayre

Frank Bernhard Schmuck

Jose Eduardo Moreira

Douglas J Griffith

Soumitra Sarkar

Yi Zhou

Christopher V Derobertis

Theodore Brian Vojnovich

Roger Lee Haskin

Shah Mohammad Rezaul Islam

Robert J Curran

Thomas K Clark

Sudhir Gurunandan Rao

Steven R Champagne

John Robert Hare

Serban C Maerean

Robert R Gensler, Jr

Hemant R Suri

John Hare

Robert Gensler

Hemant Suri

A site asymmetric topology reconciliation module (SATRM) provides a stable topology for nodes located at different sites of the cluster during loss and reconnection of communication links between the sites. The SATRM monitors the cluster topology for changes in communication links between nodes. When there is an unstable cluster topology due to a loss in the communication links, the SATRM severs links to one or more sites to create a stable topology. When a communication links recovers, the SATRM merges sites to create a stable topology with the sites connected with the recovered communication links.

Site asymmetric topology reconciliation in a computer cluster

Disclosed aspects relate to sunder management with respect to a shared pool of configurable computing resources having a cluster of disperse nodes. A set of sunder event response prioritization data may be determined based on a set of cluster profile data for the cluster of disperse nodes. A sunder event which indicates a partitioned status may be detected with respect to the cluster of disperse nodes. The set of sunder event response prioritization data may be examined. In response to detecting the sunder event, a sunder event response operation may be performed to maintain asset availability.

Sunder management for a cluster of disperse nodes

A quorum service detects liveness failures of at least two failed nodes in a domain of a cluster infrastructure layer of a cluster environment within a limited time frame and adds the at least two failed nodes to a list of nodes set to pending to be fenced by a group leader node. The quorum service determines whether the at least two failed nodes include the group leader node. The quorum service, responsive to the at least two failed nodes not including the group leader node, triggers the group leader node to trigger at least one fencing operation to fence the at least two failed nodes in the list of nodes. The quorum service, responsive to the at least two failed nodes including the group leader node, sets a new node as the group leader node and triggers the new node set as the group leader node to trigger the at least one fencing operation to fence the at least two failed nodes in the list of nodes.

Automated node fencing integrated within a quorum service of a cluster infrastructure

A system to improve communication security in cluster machine processing may include interconnected computers that can jointly process data. The system may also include a shared secret key used by each of the interconnected computers to encrypt, decrypt, and/or authenticate data being sent, or received, from one of the interconnected computers to another of the interconnected computers. The system may further include a new shared secret key used by each of the interconnected computers to encrypt, decrypt, and/or authenticate data being sent, or received, from one of the interconnected computers to another of the interconnected computers. In addition, the new shared secret key may coexist with the shared secret key without adversely affecting the joint processing of data performed by the plurality of interconnected computers.

Secure communications in computer cluster systems

A quorum service within a cluster infrastructure layer of a cluster environment comprising a plurality of nodes automatically triggers at least one automated fencing operation integrated within the quorum service, to reliably maintain a node usability state of each node of the plurality of nodes indicating an availability of each node to control and access at least one shared resource of the cluster. The quorum service reports the node usability state of each node as a cluster health status to at least one distributed application within an application layer of the cluster environment, to provide a reliable cluster health status of the plurality of nodes to the at least one distributed application for a failover of said at least one shared resource from control by a failed node from among the plurality of nodes to another node from among the plurality of nodes.

Globally consistent information is locally provided to communications layers. Globally consistent information is stored in a Network Availability Matrix, which is locally accessible by a communications layer. If an event is detected, the communications layer is automatically notified by the Network Availability Matrix, and is able to use the information in the Network Availability Matrix to quickly take action.

Locally providing globally consistent information to communications layers

The acquisition of a lock among nodes of a divided cluster is disclosed. A method is performable by each of at least one node of the cluster. A node waits for a delay corresponding to its identifier. The node asserts intent to acquire the lock by writing its identifier to X and Y variables where another node has failed to acquire the lock. The node waits for another node to acquire the lock where the other node has written to X, and proceeds where Y remains equal to its own identifier. The node waits for another node to acquire the lock where the other node has written to a Z variable, and writes its own identifier to Z and proceeds where the other node has failed. The node writes a value to Y indicating that it is acquiring the lock, and maintains acquisition by periodically writing to Z.

Lock acquisition among nodes of divided cluster

A three-dimensional data structure for storing data is defined and maintained. The three-dimensional data structure includes at least one column of data having a depth greater than zero indicating that multiple data values are provided for multiple domains represented by the structure. The data structure also includes zero or more columns of data having a depth of zero indicating that the data value of a zero depth column is invariable for a plurality of domains represented by the structure. The structure is maintained by employing various operations including, create, write and read operations.

Three-dimensional data structure for storing data of multiple domains and the management thereof

A distributed system has a multiplicity of nodes. A group of the nodes subscribe to a group services process. The group is subject to dissolving. A list of persistent subscribers to the group is maintained. Subscriptions to members and states of the group that are persistent subscribers do not terminate in response to the group being dissolved. The persistent subscribers automatically receive a notice in response to the group being reestablished and do not need to re-subscribe to the group.

Persistent group membership in a distributing computing system

The notion of controlling, using and monitoring remote resources in a distributed data processing system through the use of proxy resource managers and agents is extended to provide failover capability so that resource coverage is preserved and maintained even in the event of either temporary or longer duration node failure. Mechanisms are provided for consistent determination of resource status. Mechanisms are also provided which facilitate the joining of nodes to a group of nodes while still preserving remote resource operations. Additional mechanisms are also provided for the return of remote resource management to the control of a previously failed, but now recovered node, even if the failure had resulted in a node reset.

Scalable method of continuous monitoring the remotely accessible resources against node failures for very large clusters

A system and method for managing clustered computer resources, and more particularly very large scale clusters of computer resources by a semi-hierarchical n level, n+1 tier approach. The top level consists of the controller nodes. A first tier is defined at the top level. At a last level, at which the cluster of controlled nodes is found, a last tier is defined. Additional levels of controller or controlled nodes may exist between the top and bottom levels. At least one intermediate tier is introduced between two of the levels and comprises at least one proxy or a plurality of proxies. A proxy is a process or set of processes representing processes of the clustered computer resources. Proxies can run either on controller nodes or on the controlled nodes or controlled node clusters to facilitate administration of the controlled nodes.

Semi-hierarchical system and method for administration of clusters of computer resources

A method for managing node-specific quorum roles in a cluster of nodes. The method comprises assigning a configuration change of node role attributes to a node in the cluster, with each attribute being represented by a bit of a binary value. The binary value corresponding to the specified node role attributes is converted to an integer value and the integer value is stored in a node record associated with the node. The node record is accessed to isolate each bit of the integer value associated with the node and the node role attributes assigned to the node are identified from each of the isolated bits of the integer value associated with the node. A determination of whether the configuration change can be safely supported by a configuration quorum of nodes required to represent the cluster's configuration accurately is made based on the identified node role attributes. An additional determination of whether a required operational quorum of nodes permits access to critical and shared resources is made based on the identified node role attributes. Endorsement for the configuration change is sought from subsystems that may be affected by the configuration change. The identified node role attributes are propagated to other nodes within the cluster.

Method for managing cluster node-specific quorum roles

Growing community of inventors

Pleasant Valley, NY, United States of America

Myung Mun Bae