Up: Linkage in the Nemesis
Previous: Pervasive Interfaces
Machine addresses provide unique identifiers for interfaces
within one machine, but a higher-level naming scheme is
required for several reasons, for instance:
The important thing to realise is that these naming issues are
completely orthogonal to the idea of a single address space, and so
almost any existing name space scheme can be used. Our approach is
currently based on a directed graph of contexts referred to with
pathnames, together with a simplified version of the ANSA Trader's
constraint language (). This allows us a lot of
expressive power in the name space when we need it (and extends
naturally to the distributed case), but in well-known contexts with
few or no constraints name lookup can be extremely fast.
- We want to refer to entities in distributed systems, which have
rather different naming requirements.
- On a single machine, programs do not know addresses of other
components when they are linked. These must be determined using other
names at load time and run time.
- Human-readable names are required for users, managers
configuring the system or examining it remotely, and programmers
building and debugging it.