The highly scalable infrastructure of large-scale distributed systems is very attractive for network services. However, data access is unpredictable in this environment for the reasons of loosely coupled nature and large-scale data storage of such systems. Today, an increasing number of network applications require not only considerations of computation capacity of servers but also accessibility for adequate job allocations. An effective and adaptive mechanism of access control is important in this environment. In our study, the client clustering is used to describe the behaviors of clients and the adaptive server clustering is used to divide the large-scale distributed system into relevant small-scale systems. Since the clients which are assigned to one server cluster have the similar behaviors, we can use the stochastic control and passive measurement to do reliable and adaptive accessibility estimation and client allocation in such a small-scale system. We call this adaptive mechanism of access control based on accessibility estimation and client clustering as ACEC, and the experimental results show that ACEC can significantly reduce the data access cost and guarantee the load balance and controllability of large-scale distributed systems.