Academics and industry experts are advocating for going from large-centralized Cloud Computing infrastructures to smaller ones massively distributed at the edge of the network. However to favor the adoption of such a model, the development of a system in charge of turning such a complex and diverse network of resources into a global Cloud is critical.
In this talk, we introduce the premises of such a system. The novelty of our work is that instead of developing “yet another” brokering solution, we chose to revise the OpenStack internals with P2P mechanisms in order to operate in a distributed manner but throughout the same software platform such a geographically spread Cloud. More precisely, we describe how we extended Nova with an additional driver allowing the use of a distributed key/value store instead of the centralized SQL backend. Results of experiments conducted on Grid'5000 are promising and pave the way toward a first large-scale and WAN-wide IaaS manager based on OpenStack.
The Discovery initiative aims at operating a platform composed of thousands of servers deployed across hundreds of geographically distributed sites. Although OpenStack is organized following the Shared Nothing principle, its scalability is limited by the usage of relational databases and the use of an AMQP bus. These limitations become clearer in a multi-site scenario where synchronization between the DBs should be performed.
In this talk, participants will learn how we succeeded to replace MySQL by Redis thanks to ROME, an SQLAlchemy-like ORM that enables the use a Key/Value store system. Our implementation is promising as 80% of the API requests are completed faster than with MySQL (understanding how to improve the remaining 20% is an on-going task). Finally we will discuss results from multi-site experiments conducted on top of Grid’5000. They demonstrate the relevance of our proposal, while preserving higher-level mechanisms like host-aggregates and the usage of the Rally benchmark.