Management and Contextualization of Scientific Virtual Appliances

G Moltó and V Hernández. Management and Contextualization of Scientific Virtual Appliances. In Cloud Futures 2010: Advancing Research with Cloud Computing, 2010.

Download

(unavailable)

Abstract

The advent of Cloud computing technologies introduces new challenges for the execution of scientific applications, which typically require specific hardware and software configurations. Currently, Virtual Machine Managers (VMMs) such as Eucalyptus or OpenNebula pave the way to manage VMs in elastic Cloud infrastructures, providing moderate support for VM contextualization (e.g. IP allocation, hot-plugging disks at boot time, etc.). However, the process of turning a VM into a Virtual Appliance (VA) that encapsulates the entire hardware and software configuration for an application to run successfully is still fairly manual. Scientific applications typically rely on numerical libraries, databases, web services, system packages, etc. whose installation and configuration can sometimes be automatically performed. Therefore, we envision semi-automatic procedures to contextualize VMs for scientific applications, where users and/or developers would provide a declarative description of their applications and its dependences. Then, contextualization software would resolve the dependences and provide the appropriate environment to install each requisite. However, installing complex software might not be automated so easily. This would lead to a set of Pre-Contextualized VMs (PCVMs) where the user (or the VM provider) would have installed this kind of software. These PCVMs could be stored in VM providers (such as Amazon S3) and a VM Repository Service would store metadata information about each VM. This would allow PCVMs sharing among different scientists, thus enhancing collaboration. These PCVMs would certainly be finally configured at boot time to produce a self-contained VA that allows the execution of the application. We are currently working towards this vision and have developed Proof-of-Concept tools that automatically deploy scientific applications that rely on different build systems (ant, make), different software packages (Tomcat, Globus Toolkit WS Core), and different configuration files (XML, INI files). These tools, although in their early stages, ease the deployment of applications onto base virtual machines.

BibTeX Entry

@inproceedings{Molto2010mcs,
   abstract = {The advent of Cloud computing technologies introduces new challenges for the execution of scientific applications, which typically require specific hardware and software configurations. Currently, Virtual Machine Managers (VMMs) such as Eucalyptus or OpenNebula pave the way to manage VMs in elastic Cloud infrastructures, providing moderate support for VM contextualization (e.g. IP allocation, hot-plugging disks at boot time, etc.). However, the process of turning a VM into a Virtual Appliance (VA) that encapsulates the entire hardware and software configuration for an application to run successfully is still fairly manual. Scientific applications typically rely on numerical libraries, databases, web services, system packages, etc. whose installation and configuration can sometimes be automatically performed. Therefore, we envision semi-automatic procedures to contextualize VMs for scientific applications, where users and/or developers would provide a declarative description of their applications and its dependences. Then, contextualization software would resolve the dependences and provide the appropriate environment to install each requisite. However, installing complex software might not be automated so easily. This would lead to a set of Pre-Contextualized VMs (PCVMs) where the user (or the VM provider) would have installed this kind of software. These PCVMs could be stored in VM providers (such as Amazon S3) and a VM Repository Service would store metadata information about each VM. This would allow PCVMs sharing among different scientists, thus enhancing collaboration. These PCVMs would certainly be finally configured at boot time to produce a self-contained VA that allows the execution of the application. We are currently working towards this vision and have developed Proof-of-Concept tools that automatically deploy scientific applications that rely on different build systems (ant, make), different software packages (Tomcat, Globus Toolkit WS Core), and different configuration files (XML, INI files). These tools, although in their early stages, ease the deployment of applications onto base virtual machines.},
   author = {G Moltó and V Hernández},
   booktitle = {Cloud Futures 2010: Advancing Research with Cloud Computing},
   title = {Management and Contextualization of Scientific Virtual Appliances},
   year = {2010}
}