RDMA server (OSI) global TCE tables
View Patent ↗In remote direct memory access (RDMA) transfers in a multinode data processing system in which the nodes communicate with one another through communication adapters coupled to a switch or network, there is a need for the system to ensure efficient memory protection mechanisms across jobs. A method is thus desired for addressing virtual memory on local and remote servers that is independent of the process ID on the local and/or remote node. The use of global Translation Control Entry (TCE) tables that are accessed/owned by RDMA jobs and are managed by a device driver in conjunction with a Protocol Virtual Offset (PVO) address format solves this problem.
1. A method for providing remote direct memory access data transfers in a multinode data processing system in which the nodes communicate with one another through communication adapters coupled to a switch or network, said method comprising the steps of:
establishing, in operating system address space within said nodes, a set of node resident control tables with each table having entries which include a real address and a key which links said table entries to respective ones of task within user address spaces within said nodes, said real address pointing to a data buffer transfer area within said user address space;
establishing, in memory within said communication adapters, an adapter resident control table, corresponding at least in part to said node resident control tables, which includes an address for said node resident control table and a job identifier; and
transferring, within said multinode system, data packets having header information which includes a job key associated with said task, a control table identifier and data offset information.
2. The method of claim 1 further including a step of using said control table identifier to access a job identifier within said resident adapter control table and comparing said job identifier with a job key provided within said header.
3. The method of claim 1 in which said control tables are shared across a plurality of tasks running within a single job.
4. The method of claim 1 in which said control tables provide remote direct memory access to select ones of said jobs.
5. The method of claim 1 in which access to said tables is limited so that no table is accessible by more than one job.
6. The method of claim 1 in which said translation control tables are established by a device driver running on its respective node.
7. The method of claim 1 in which said operating system is a hypervisor.
8. The method of claim 1 in which said data buffer is made addressable to all adapters on the node that are active for a job associated with said job key.
9. The method of claim 1 in which said adapter resident control table includes a table size.
10. The method of claim 1 in which said adapter resident control table includes a page size.