Micro Kernel Virtualization for High Performance Cloud and HPC systems (MIKELANGELO)

Micro Kernel Virtualization for High Performance Cloud and HPC systems

Project Goals

The European-Israeli project MIKELANGELO aims to fundamentally disrupt the architecture of virtual infrastructures for cloud computing. The project covers the whole software range of the modern computing stack for a broad set of use cases. Up to now compute-intensive applications such as big data and high-performance computing (HPC) do not use cloud services by default, since the input/output performance (I/O) is too low. If the I/O performance is not nearly as high as in the case of native installations the costs increase to purchase additional or more powerful hardware. cloud computing relies heavily on virtual machines (VMs). The physical hardware (server, storage systems, network technology) builds the foundation and on top of this lies the virtualisation layer. It offers near-zero overhead for computation, however reaching only about 60-70% efficiency for I/O operations.

This is the starting point of the MICHELANGELO project. The members work to improve the I/O performance of virtualized infrastructures and applications running in cloud. The consortium targets to increase I/O efficiency to nearly 100%, in an improved hypervisor, called sKVM. sKVM will be integrated with Remote Direct Memory Access (RDMA), which allows direct access from the memory of one into the memory of another one with a minimum of communication overhead. No additional CPU overhead in VM is incurred, and I/O in VMs is increased. Improving sKVM is one of the key features of the project. The operating system OSv and the tool Capstan build on this feature to enable agile deployment of applications. Both are developed by Scylla DB, one of the MIKELANGELO project members.

MIKELANGELO ’s architecture also will include functionalities to ensure security in the host OS, hypervisor, and in the cloud middleware.

GWDG’s Role in the project

The GWDG integrates both sKVM and OSv with OpenStack, to provide the advancements in a productive environment to users of infrastructure services. This integration with OpenStack will include a new application deployment model, based on OSv. The GWDG embeds Capstan in the cloud layer, to deploy a large number of applications conveniently via various user interfaces. In addition, the cloud layer will feature advanced monitoring across all layers of the architecture, even for user-deployed applications.

Examples of use cases:

Big data cluster-on-demand
Cloud bursting (on demand outsourcing of applications into a public cloud if extra resources are required.)
HPC with VMs, as a large-sized real-time application in the cloud

At the end of the project GWDG aims to offer big data services on demand to its customers.