We describe how to control the cumulative use of distributed grid resources by using coordination-aware policy decision points (coordinated PDPs) and an SQL database to hold 'coordination' data. When access to a resource is granted, obligations in the security policy ensure that the coordination database is updated. The coordination database is a normal grid service providing distributed access to the coordinated PDPs. Access to the databases is secured by the grid security infrastructure (GSI) and its own PDP, so that only authorized users (the coordinated PDPs) can access it. A coordinated PDP is imbedded into the Globus Toolkitv4 authorization chain as a custom PDP so that any grid service can be protected by a security policy that provides a coordination capability. Each coordinated PDP uses the services of an uncoordinated PDP to make its access control decisions, so that any existing stateless PDP can be supplemented with a coordination capability. We provide performance results for the coordinated PDPs and compare these with two stateless PDPs. Virtually the entire performance penalty of using coordinated PDPs is accounted for by the heavy costs of using GSI to secure communications between the coordinated PDPs and the coordination database. around with him. Providing a similar capability for grid jobs, for example, to limit the amount of storage or cpu that a user may request per day or per job from any location on the grid, is not so easy. The grid job will almost certainly run on different machines under different administrative control, will probably run under different account names on each machine, and the access control mechanism of one machine is typically unable to communicate with those of the other machines. The security token that is often passed from machine to machine is the proxy certificate [1], but this is not used by the policy decision points (PDPs) to communicate with each other and is not under their direct control (unlike the bank card inserted into an ATM). Consequently, the design of a policy-based coordinated access control system presents a number of challenges.The lack of communication between the PDPs of distributed applications can be addressed today by sidestepping the issue and using a centralized PDP with a common policy that is used by all the grid resources (see Figure 1). Such a system has been available for several years to grid applications that use Globus Toolkit (GT) from v3.3 onwards. GT is capable of making an external authorization callout using the GGF SAML authorization protocol specification [2], and several PDPs such as PERMIS [3] and PRIMA [4] have implemented this protocol. This sort of access control infrastructure allows a common policy to be used by all the resources of a grid, but since most PDPs today are stateless they are still unable to coordinate their access control decisions across multiple access requests. A further disadvantage of this configuration is that the central PDP is a bottleneck to performance because every request needs to be diverted ...