APPLICATION-AWARE MULTI-TENANT CoFlow SCHEDULING MECHANISM

Distributed parallel computing frameworks are widely-used in multi-tenant data centers, where applications from different tenants share and compete for bandwidth resources. Distributed parallel computing frameworks mostly use Coflow as the traffic model. In order to solve the problem of performance optimization and tenant isolation in the Coflow scheduling process, an application-aware multi-tenant Coflow scheduling mechanism, Aamt, is proposed. According to the tenant isolation problem, the progress was introduced as a measurement standard. According to the dynamic and elastic changes of tenant needs, the tenant arrival model was designed to obtain the long-term progress guaranteed by tenant isolation, so as to achieve application awareness. In order to improve the network bandwidth utilization, the Lagrange dual optimization was designed to obtain performance optimization progress. The experimental results show that Aamt can not only ensure tenant isolation, but also increase bandwidth and improve link utilization.