Large cloud companies manage dozens of datacenters across the globe connectedusing dedicated inter-datacenter networks. An important application of thesenetworks is data replication which is done for purposes such as increasedresiliency via making backup copies, getting data closer to users for reduceddelay and WAN bandwidth usage, and global load balancing. These replicationsusually lead to network transfers with deadlines that determine the time priorto which all datacenters should have a copy of the data. Inter-datacenternetworks have limited capacity and need be utilized efficiently to maximizeperformance. In this report, we focus on applications that transfer multiplecopies of objects from one datacenter to several datacenters given deadlineconstraints. Existing solutions are either deadline agnostic, or only considerpoint-to-point transfers. We propose DDCCast, a simple yet effective deadlineaware point to multipoint technique based on DCCast and using ALAP trafficallocation. DDCCast performs careful admission control using temporal planning,uses rate-allocation and rate-limiting to avoid congestion and sends trafficover forwarding trees that are carefully selected to reduce bandwidth usage andmaximize deadline meet rate. We perform experiments confirming DDCCast'spotential to reduce total bandwidth usage by up to $45\%$ while admitting up to$25\%$ more traffic into the network compared to existing solutions thatguarantee deadlines.
展开▼
