Microsoft Word - review_response_OLP_2.docx

We th p L (i) Fr c req A w (ii) A an r (2010), s l T th T c w h byd c v B f c r c red a m B al as m b al m l -s ank the reviewers for the careful feedback and appreciate the time spent reading our aper. Detailed responses are as below: iterature on online LP ( OLP) and the contribution of our work: om the algorithmic perspective, our algorithm has a strongly polynomial O ( nnz( A)) flop omplexity ( linear in the number of non-zero entries in A), while the previous OLP algorithms all uire solving O ( log n) or O ( n) of LPs (increasing to the full size over time). For example, grawal et al. ( 2 0 1 4) solved O ( log n) LPs and Kesselheim et al. (2014) solved O(n) LPs. As far as e know, the algorithm is the first of its kind and the most efficient OLP algorithm so far. As mentioned by the eviewer, our algorithms share similarity with the network control algorithm in Neely, M. J. but our analysis extends their analysis ( in i. i.d.