Next: Real-time Computations Up: Partially Ordered Runs Previous: Partially Ordered Runs

Runs

For simplicity, we restrict attention to pure runs and deterministic agents. A run rho of a distributed ealgebra A can be defined as a triple (M, A, sigma) satisfying the following conditions 1-4.

1.
M is a partially ordered set, where all sets {y: y leq x} are finite.

Elements of M represent moves made by various agents during the run. If y < x then x starts when y is already finished; that explains why the set {y: y leq x} is finite.

2.
A is a function on M such that every nonempty set {x : A(x) = a} is linearly ordered.

A(x) is the agent performing move x . The moves of any single agent are supposed to be linearly ordered.

3.
sigma assigns a state of A to the empty set and each initial segment of M ; sigma(emptyset) is an initial state.

sigma(X) is the result of performing all moves in X .

4.
The coherence condition: If x is a maximal element in a finite initial segment X of M and Y = X - {x}, , then A(x) is an agent in sigma(y) and sigma(X) is obtained from sigma(Y) by firing A(x) at sigma(Y) .

Intuitively, a run can be seen as the common part of histories of the same computation recorded by various observers. We hope to address this issue elsewhere.

If agents are not necessarily deterministic, we have to define moves as state transformers and make the coherence condition more precise:

4'
If x is a maximal element in a finite initial segment X of M and Y = X - {x}, then A(x) is an agent in sigma(Y) , x is a move of A(x) and sigma(X) is obtained from sigma(Y) by performing x at sigma(Y) .

A run rho' is an initial segment of a run rho if (i) the move poset of rho' is an initial segment of the move poset of rho and (ii) the agent and state functions of rho' are restrictions of those in rho . A run rho' is a linearization of rho if the move poset of rho' is a linearization of that of rho , the agent function of rho' is that of rho , and the state function of rho' is a restriction of that of rho . Linearizations are sequential runs. A state S of is reachable in a run rho if it belongs to the range of the state function of rho .

Corollary 6.1. All linearizations of the same finite initial segment of rho have the same final state.

Corollary 6.2. A property holds in every reachable state of a run rho if any only if it holds in every reachable state of every linearization of rho.



Next: Real-time Computations Up: Partially Ordered Runs Previous: Partially Ordered Runs


huggins@acm.org
Thu Mar 23 17:30:35 EST 1995