Resource reservation and scheduling in the IaaS model
The IaaS model is a cloud computing paradigm where the provider gives a direct access to virtual resources (network, storage, virtual machines, …) on which the client can install his own services and maintains them completely on his own. Today there are many companies providing commercial solutions for this type of service, like Amazon EC2 or Google Cloud Compute.
These companies usually use their own software to manage their infrastructure, but there also Open Source softwares like OpenNebula or OpenStack which can be installed on private clusters to provide a private IaaS cloud to, for instance, a company's users.
One of the major difficulty is the resource allocation, that is how the available physical resources should be managed so that they are used in an optimal manner.
There exists a number of scheduling algorithms who manage the requests and distribute them across various resources, in our case it is necessary to include an advance reservation part, how do these algorithms perform with multiple reservations at multiple, overlapping time slots ? In the framework of the European project Contrail, we integrate a Service Level Agreement (SLA) mechanism which adds complexity to the allocation task by introducing new constraints like the location of virtual machines (All on the same hosts, on separate hosts, in a specific country …), these constraints may be used by the client for security or reliability reasons. It is also interesting to have a preemption mechanism, introducing a way for user's applications to have a set priority, it would for instance allow an HPC application to take over nodes used by other lower priority applications, stopping them and resuming them at a later date.
Tasks
The intern will study the different existing scheduling algorithms and their applicability to our needs, specifically their interaction with SLA constraints and applications' priority, in the framework of an advance reservation mechanism.
He will then be tasked with the design and implementation of a module which will be integrated in the Contrail software Inria is developing.
Another task, if there is interest and time, could be the study of Pricing models and their integration with our reservation module.
How to apply
Inviare il proprio CV in inglese attraverso questo FORM il prima possibile.
Per maggiori informazioni.
The IaaS model is a cloud computing paradigm where the provider gives a direct access to virtual resources (network, storage, virtual machines, …) on which the client can install his own services and maintains them completely on his own. Today there are many companies providing commercial solutions for this type of service, like Amazon EC2 or Google Cloud Compute.
These companies usually use their own software to manage their infrastructure, but there also Open Source softwares like OpenNebula or OpenStack which can be installed on private clusters to provide a private IaaS cloud to, for instance, a company's users.
One of the major difficulty is the resource allocation, that is how the available physical resources should be managed so that they are used in an optimal manner.
There exists a number of scheduling algorithms who manage the requests and distribute them across various resources, in our case it is necessary to include an advance reservation part, how do these algorithms perform with multiple reservations at multiple, overlapping time slots ? In the framework of the European project Contrail, we integrate a Service Level Agreement (SLA) mechanism which adds complexity to the allocation task by introducing new constraints like the location of virtual machines (All on the same hosts, on separate hosts, in a specific country …), these constraints may be used by the client for security or reliability reasons. It is also interesting to have a preemption mechanism, introducing a way for user's applications to have a set priority, it would for instance allow an HPC application to take over nodes used by other lower priority applications, stopping them and resuming them at a later date.
Tasks
The intern will study the different existing scheduling algorithms and their applicability to our needs, specifically their interaction with SLA constraints and applications' priority, in the framework of an advance reservation mechanism.
He will then be tasked with the design and implementation of a module which will be integrated in the Contrail software Inria is developing.
Another task, if there is interest and time, could be the study of Pricing models and their integration with our reservation module.
How to apply
Inviare il proprio CV in inglese attraverso questo FORM il prima possibile.
Per maggiori informazioni.
No comments:
Post a Comment