Term
|
Definition
|
Cloud Provisioning Environment (CPE)
|
The name of collaboration project between UCSD and ANL (during the OOI pilot period, 2009) to develop a prototype for an elastic scaling cloud execution architecture. The prototype covers distributed architecture elements that communicate via a message-passing infrastructure in a pub/sub style, necessary interaction specifications and message formats, integration of existing elastic provisioning and computing services, and development of elastic provisioning policy
|
Common Execution Infrastructure (CEI)
|
One of the six subsystems of the OOI CI. It is based on the Common Operating Infrastructure (COI) subsystem and its services. The CEI provides services for flexible execution of processes in multiple execution environments. One of its services is elastic provisioning. It has further services not covered in this prototype.
|
Sensor (elastic computing)
|
In this context, a software process in the network that emits a sensor signal to a sensor aggregator. Common sensors measure CPU load, disk I/O utilization, network bandwidth, work queue length, state of health information, etc.
|
Sensor Aggregator
|
Receives sensor signals through various channels (TCP, SMNP, ...) and aggregates them into events delivered via the message-passing system
|
Scheduler
|
Also known as actuator. Based on currently active elastic computing policy and received events from the sensor aggregators; decides on provisioning and taking down computing resources. Emits actions targeted at provisioners and transported via the message-passing system.
|
Provisioner
|
Also known as deployment orchestrator. Process that acts on provisioning actions (commands) and creates and takes down new computing instances. Its implementation is specific for each supported target execution environment, such as EC2 and Nimbus.
|
Launcher
|
Part of the provisioner. Also known as factory. Creates a new execution resource, such as an AMI instance, by accessing the interfaces of a cloud resource provider, such as the AWS. Provides initial metadata to the resource to be creaed via the native mechanism, for instance the network address of the context broker via the Amazon EC2 metadata key/value pairs.
|
Context Broker
|
Part of the provisioner. Service available at a know location in the network that is accessible by a newly created (launched) execution resource, such as an AMI instance. The context broker receives the identification of the new instance and provides it with all required context for its operation. Context includes network name, metadata, configuration, applicable updates etc.
|
Policy (elastic computation)
|
Configuration information and rules determining which resources to provision or take down where and with which parameters.
|
Elastic Processing Unit (EPU)
|
A distributed application identified by a single unique name. An EPU consists of a work queue, multiple instances of worker processes that feed from the queue and management mechanisms to bring up and control the instances and the EPU itself. Not within the scope of the CPE prototype.
|
Message-passing system
|
In this prototype, a RabbitMQ broker implementing the AMQP 0.8 specification. Clients are implemented in Python based on the Twisted framework. The UCSD Magnet framework realized the Twisted internal protocol adapter for AMQP, such that clients are isolated from the AMQP interface.
|
Execution Environment
|
Such as Amazon EC2 or an installation of Nimbus. Provides a specific interface to create and control execution resources, such as compute nodes (AMI instances) and storage nodes (EBS volumes).
|
Page Operations
Browse Space