In modern data centers based on virtualization, virtual machine (VM) allocation is the primary consideration for efficient resources scheduling in the cloud. It has been proved that the VM allocation (VMA) problem to minimize the maximum access latency is of NP-hard. But few works on fault tolerant algorithms for the VMA problem are reported, which is very important for the security and reliability of the data center network. We present a heuristic algorithm focusing on minimizing the maximum data access latency. The proposed fault tolerant algorithm constructs the group with controllable redundancy in the set of available VMs, and then selects the VMs from the group and assigns them to the data nodes in the data center network. Extensive experiments on the Tree, VL2, Fat-Tree and BCube, the four commonly used network structures, show that the proposed algorithm can provide between 0 and 200% redundancy, the assignment time of virtual machines for the data nodes is reduced by 67.1% on average when redundancy is between 0 and 40%, and the running time of the algorithm is generally reduced by 12.8%.