With　the　arrival　of　the　5G　era,　a　new　service　paradigm　known　as　mobile-edge　computing　(MEC)　has　been　introduced　for　providing　high　quality　mobile　services　by　offloading　the　delay-sensitive　and　computation-intensive　tasks　from　mobile　devices　to　nearby　MEC　servers.　In　this　paper,　we　investigate　the　problem　of　delaysensitive　task　scheduling　and　resource　(e.g.　CPU,　memory)　management　on　the　server　side　in　multi-user　MEC　scenario,　and　propose　a　new　online　algorithm　based　on　deep　reinforcement　learning　(DRL)　method　to　reduce　average　slowdown　and　average　timeout　period　of　tasks　in　the　queue.　We　also　design　a　new　reward　function　to　guide　the　algorithm　to　learn　directly　from　experience　to　scheduling　tasks　and　managing　resources.　Simulation　result　shows　that　our　algorithm　outperforms　multiple　traditional　algorithms　and　have　a　big　advantage　of　intelligence　and　good　understanding　towards　workload　and　environment.　©　2019　Association　for　Computing　Machinery.