(Français) Projet DiVa

Disaggregated Virtualization

— Preview

Study of new virtualisation mechanisms adapted to a disaggregated cloud infrastructure and to a cloud infrastructure made up of small edge infrastructures connected to powerful data centres.

Gaël Thomas, Inria Senior Researcher

Keywords : Disaggregation, edge, virtualisation

Data centers are evolving with the arrival of disaggregated and edge infrastructures. The aim of the DiVA project is to prepare the system stack for these next-generation infrastructures, in order to make more efficient use of hardware resources.
 
In the context of disaggregated infrastructures, it is becoming possible, on the scale of a cluster of machines, to fully mutualize hardware resources. This evolution requires us to revisit virtualization mechanisms, as virtual machines become elastic (new resources can be added to a virtual machine on the fly) and distributed (the hardware resources used by a virtual machine are physically distributed).
 
In the context of edge infrastructures, small clusters of machines are connected by slow networks to powerful data centers. Edge clusters perform computations close to data sources, while still being able to use the computing power of a data center. This evolution requires us to revisit several system mechanisms, as infrastructures become much more heterogeneous in terms of hardware, network and software architecture.

— Missions

— Our researches

System mechanisms for a disaggregated data center

Propose new system mechanisms to make better use of a disaggregated infrastructure.

Define new virtualization interfaces to enable a virtual machine to dynamically allocate or release hardware resources,

Investigate new scheduling and placement mechanisms in the guest operating system to make efficient use of these distributed resources,

Investigate how we can use programmable networks to optimize virtual machine performance.

Study how we can design transparent replication mechanisms since, in this distributed context, faults are no longer exceptions but become the norm.


Continuum edges – data centers

Propose new system mechanisms to better manage the edge/cloud continuum.

Design new virtual machine migration mechanisms to migrate virtual machines between heterogeneous machines (instruction set and hypervisor).

Seamlessly optimize the data paths of distributed multi-tier applications between cloud and edge to optimize the edge/cloud network.
Define new virtualization interfaces to optimize short-lived virtual micro-machines.

— Partners

Consortium

Inria, GINP, Télécom SudParis, Toulouse INP, Sorbonne Université

— Research team

9 permanent staff (not funded by the project)
16 people funded by the project, including :
8 PhD students
8 post-docs

Our teams in France


Other projects

Comments are closed.