Implementation of HSM under DFS

Using the provided DFS form The Open Group, the RPC time out occurs because the handling, mounting and loading of a tape takes several minutes in the worst case. In the specification of the DFS, the client disconnects the RPC connection to the server unless the client receives a response within a finite time (4.5 minutes).

To avoid this RPC time-out, we added the following mechanism on both Hitachi's DCE and the HSM, OMNI. For example, we will explain the control flow for the `READ' action (See Fig. 3). The `WRITE' action is similar to it.

1. Application requests the `READ' action to the DFS server through the DFS client.
2. If the file exists on disks, data transmission starts immediately, as usual.
3. If not, the DFS server sends a request for migration to the OMNI client. Then, the OMNI client sends a asynchronous request of migration-to-disk to the OMNI server, and sends return code with E_NODISC to the DFS client through the DFS server.
4. The DFS client repeatedly trys to send a request until it receives the normal return code.

Figure 3: Mechanism of DFS with OMNI