Tabu Search
• Heuristic search method for fast and
nearly optimal solving of optimization
problems
 Starting from an initial solution, Tabu
Search picks either the best better
neighbor solution or the neighbor
solution with least degradation of result
• Loops are prevented by considering
only solutions which haven't been
considered before
 Requires temporary storing of
last n solutions (TabuFifo)
HW/SW-Codesign / System Synthesis - 1
• Escapes from local minima
• Accepting a new solution implies
removal of oldest solution from
TabuFifo (if TabuFifo is full)
• Length of TabuFifo influences
effectiveness of method
 TabuFifo too small: Loops may occur
 TabuFifo too large: Possibly no new
neighbors are found which weren’t
considered yet
© Lehrstuhl für Integrierte Systeme
Tabu Search Algorithm

CurrentSolution:=InitialSolution • Initialization
BestSolution:= CurrentSolution
TabuFifo := { } // Size nTabu_Fifo • Pick best solution from neighborhood
REPEAT which isn‘t yet stored in TabuFifo
N := Neighbourhood(CurrentSolution) • Accept degradation in result.
NewSolution := NoSolution
FOR all Elements X  N AND X  TabuFifo
IF (cost(X) < cost(NewSolution)) THEN • No further solution possible,
NewSolution := X i.e. N  TabuFifo
ENDIF • Stop Tabu Search
ENDFOR
IF (NewSolution = NoSolution) THEN
• Select solution for next iteration
EXIT • Update Tabu-Fifo
CurrentSolution := NewSolution • Memorize best solution found so far
Enter_in_TabuFifo(CurrentSolution)
BestSolution := Select_MinCost(BestSolution, CurrentSolution )
• Stop Tabu Search with
UNTIL (stopcondition) other stop criterium cost(NoSolution) = 

HW/SW-Codesign / System Synthesis - 2 © Lehrstuhl für Integrierte Systeme

 nTabu_Fifo = 5
Tabu Search - Example Termination condition:
4 x no improvement in BS

BS: CS: Step:

L1 L2 L3 L4 L5 L6 L7 L8 L9 L10 L11 L12
99 99 Start: L27
L13 L14
L15 L16 L17 L18 L19 L20 L21 L22 L23 L24 91 91 L27 L38
C=98
C=99 C=97 89 89 L38 L51
L25 L26
L27 L28 L29 L30 L31 L32 L33 L34 L35 L36
C=97 C=95
C=99 C=98 85 85 L51 L63
L37 L38
L39 L40 L41 L42 L43 L44 L45 L46 L47 L48 79 79 L63 L76
C=95 C=91
C=95 C=98 C=69 C=65 C=61 79 80 L76 L77
L49 L50
L51 L52 L53 L54 L55 L56 L57 L58 L59 L60
C=93 C=92
C=89 C=94 C=71 C=65 C=59 C=65 79 81 L77 L89
L61 L62
L63 L64 L65 L66 L67 L68 L69 L70 L71 L72 79 84 L89 L90
C=91
C=85 C=87 C=85 C=89 C=74 C=65 C=60 C=64 C=69 77 77 L90 L79
L73 L74
L75 L76 L77 L78 L79 L80 L81 L82 L83 L84
C=84
C=82 C=79 C=80 C=84
C=89
C=99 C=77 C=71 C=62 C=63 C=67 65 65 L79 L68
L85 L86
L87 L88 L89 L90 L91 L92 L93 L94 L95 L96 59 59 L68 L57
C=85 C=86 C=81 C=85
C=84 C=91 C=80 C=73 C=65 C=61 59 60 L57 L69
L97 L98 L99 L100 L101 L102 L103 L104 L105 L106 L107 L108
C=88 C=89 C=87
C=97 C=95 59 62 L69 L81
L109 L110 L111 L112 L113 L114 L115 L116 L117 L118 L119 L120 59 63 L81 L82
59 61 L82 L95

Current solution Neighborhood (w/o Tabu Fifo content) Tabu Fifo content
HW/SW-Codesign / System Synthesis - 3 © Lehrstuhl für Integrierte Systeme