AtlasPublicCloudMagellan

Introduction

We are building a fully functional ATLAS PanDA cluster on IaaS Cloud. Currently computing resources is mainly provided by the Magellan Cloud Research project, however, the same methodology can be used on any IaaS cloud.

Comparing to a conventional PanDA site, this "cloudy" cluster has several highlights:

1. Auto-Scaling
2. Storage on Worker
3. VCA concept and CloudCRV for deployment automation
4. High Speed Link to ATLAS Data

List of Contributors

Sub-projects and Responsibilities

Mail List

atlas-cloud@googlegroupsNOSPAMPLEASE.com http://groups.google.com/group/atlas-cloud

To Join email yao.yushu@gmailNOSPAMPLEASE.com

Meetings

UML Diagrams

RoleDefinitions:

• Each RoleDef_Lead (Darker Green) corresponds to a VM when deployed.
• Each RoleDef_Lead depends on a number of RoleDef_Auto (Lighter Green), they are connected by blue dependency arrows (A->B means A depends on B).
• Each RoleDef_Auto can depend on other RoleDef_Auto in an remote VM. Shown by red arrows.
• The whole dependency graph should be a DAG (directed graph without loop).
• The RoleDef_Scale (Lighter Blue) is a special kind of RoleDef_Lead that can have more than one instances when deployed. The number of deployed instances for this RoleDef_Scale is controlled by a RoleDef_ScaleCtrl (Darker Blue) which will give feedback to CloudCRV who controls the lifetime of all VMs. A RoleDef_ScaleCtrl is a special type of RoleDef_Auto.

• Resource Pool with on Private IP Address:
  

• Resource Pool with Public and Private IP Addresss:
  

Responsible: YushuYaoLBL
Subject: grid