Server consolidation is the hosting of multiple tenants on a server machine. Given a sequence of data analytics tenant loads defined by the amount of resources that the tenants require and a service-level agreement (SLA) between the customer and the cloud service provider, significant cost savings can be achieved by consolidating multiple tenants. Since server machines can fail causing their tenants to become unavailable, service providers can place replicas of each tenant on multiple servers and reserve capacity to ensure that tenant failover will not result in overload on any remaining server. We present the CUBEFIT algorithm for server consolidation that reduces costs by utilizing fewer servers than existing approaches for data analytics workloads. Unlike existing consolidation algorithms, CUBEFIT can tolerate multiple server failures while ensuring that no server becomes overloaded. Through theoretical analysis and experimental evaluation, we show that CUBEFIT is superior to existing algorithms and produces near-optimal tenant allocation when the number of tenants is large. Through evaluation and deployment on a cluster of 73 machines as well as through simulation studies, we experimentally demonstrate the efficacy of CUBEFIT.
