RAS: Reliable auto-scaling of virtual machines in multi-tenant cloud networks

摘要

This paper addresses the problem of auto-scaling cloud services (tenants) in a failure prone cloud data center network. Auto-scaling enables existing tenants to modify their services or service availability, for example, to meet market demand fluctuations or to prepare for high-traffic events and scheduled load changes. Cloud operators, in response, need to accommodate such requests by provisioning additional resources or reconfiguring current ones. This manuscript shows that such a decision problem, in addition to being combinatorially complex, is not intuitive, highlighting several tradeoffs between disruption times due to service migration and resource redundancy to achieve higher service availability. We present RAS, an availability-aware auto-scaling scheme for managing Virtual Network (VN) scaling requests. RAS provides a robust adaptive approach for managing requests change over time while maintaining their availability requirement. Our numerical results prove that our suggested approach harnesses the benefits of migration and redundancy to achieve an adaptive scaling technique that outperforms migration or redundancy oblivious techniques.