Max progressive network update

Datacenter network always needs to be updated so that it can adapt to frequent traffic changes and constantly achieve high network utilization. However, the quality of service can be easily reduced by unexpected transient link congestion, which may be caused by variable link transmission latency during flow migration. Furthermore, when network administrators reallocate bandwidth resource without taking flow path information into consideration, deadlocks among flows waiting for link resources unavoidably block the updating process. In this paper, we prove it is NP-hard to find a feasible updating plan with the minimum number of rate-limiting flows when deadlocks occur. We propose a buffer-assisted schedule, Max Progressive Updating Method (MAPUM), to update network in a consistent and efficient way. Extensive simulations prove that MAPUM accomplishes network updating in a lossless and effective way.