Challenging the Computational Metaphor: Implications for How We Think

Actor-Based Parallel Dataflow Analysis

“…E-IFDS call-site code def CallSiteActor(n) [1] local PathEdge := ∅ [2] local CallEdge := ∅ [3] local CalleePathEdge := ∅ begin (message) switch [4] case message matches AddPathEdge d1, d2 : [5] if d1 → d2 / ∈ PathEdge then [6] Insert d1 → d2 into PathEdge [7] for each p ∈ calledProcs(n) and d3 ∈ flow call (n, d2, p) do [8] Propagate(sp, AddPathEdge d3, d3 ) [9] Insert p, d2 → d3 into CallEdge [10] for each d4 such that p, d3 → d4 ∈ CalleePathEdge do [11] for each d5 ∈ flowret(ep, d4, n, d2) do [12] Propagate(returnSite(n), AddPathEdge d1, d5 ) [13] od [14] od [15] od [16] for each d3 ∈ flow thru (n, d2) do [17] Propagate(returnSite(n), AddPathEdge d1, d3 ) [18] od [19] fi [20] case message matches AddCalleePathEdge p, d1, d2 :…”

“…is not empty then [9] Remove next (a, m) from Q[qid] [10] Release (L[qid]) [11] if TryAcquire (a.lock) then [12] a.qid := qid [13] a.execute (m) [14] Release (a.lock) [15] return true [16] else [17] Send (a, m) [18] fi [19] else lock := false end Algorithm 5: The Task-Throwing Scheduler a random queue (line 4) before checking its own queue again. Each queue is protected by a lock (from array L).…”

“…case n is not a Call Site or Exit node : N A [n] := new IntraActor(n) end case [5] end switch od [6] Tracker := new TrackerActor(currentT hread) [7] Propagate(smain, AddPathEdge 0, 0 ) [8] Wait for Done [9] return all distinct n, d2 where some d1 → d2 ∈ N A [n].PathEdge end pure function Propagate(n, message) begin [10] Send synchronous Inc to Tracker [11] Send message to N A [n] end def TrackerActor(receiver) [12] local count := 0 begin (message) switch [13] case message matches Inc : [14] count := count + 1 [15] case message matches Dec : [16] count := count -1 [17] if count = 0 then Send Done to receiver fi end Node-actor operation is identical to the corresponding operations in E-IFDS, with one exception: IFDS-A's exit node-actor forwards its edges to the call-site node-actor instead of generating summary edges itself. The reason for this is that the summary edges generated in response to facts at the exit node require access to the PathEdge and CallEdge sets present in the CallSiteActor.…”

“…Constructing this type of algorithm using traditional thread-and-lock expressions can be a difficult exercise because it requires reasoning about shared data consistency in the presence of non-deterministic thread interleavings, reasoning which is extraordinarily difficult for human minds [9,16]. We approach the task by expressing the algorithm using the Actor model [1,6].…”

Introduction to the strategy and methods of complex systems

“…Computers have made a transition from systems with tightly controlled inputs and outputs to networks that respond on demand as interactive information systems (Stein 1999). This has changed radically the nature of their design.…”