To　optimize　uplink　random　access　procedure,　improve　resource　utilization　and　allocate　limited　radio　resources　reasonably　of　Machineto-　Machine　(M2M)　communication　in　cellular　networks,　we　propose　a　novel　two-level　controlling　framework　for　M2M　communications　in　software　defined　cellular　networks　with　wireless　network　virtualization.　In　our　proposed　framework,　substrate　physical　M2M　network　is　abstracted　and　sliced　into　multiple　virtual　M2M　networks,　each　is　controlled　by　a　virtual　SDN　controller.　By　decoupling　control　plane　from　the　virtual　base　station　and　by　the　interaction　between　virtual　controller　and　virtual　equipments,　random　access　procedures　can　be　optimized　and　throughput　can　be　increased.　Furthermore,　we　propose　a　modified　particle　swarm　optimization-simulated　annealing　(PSO-SA)　algorithm,　named　temperature　in　SA　changing　inertia　weight　PSO　algorithm　(TCIWPSO-SA),　to　allocate　virtual　radio　resources　reasonably　in　order　to　maximize　fairness　among　virtual　networks.　Extensive　simulation　results　with　different　system　parameters　are　presented　to　show　that　the　proposed　scheme　and　algorithm　can　achieve　considerable　performance　gains　in　both　throughput　and　fairness　of　virtual　resource　allocation.