278 European Journal of Operational Research 64 (1993)278-285

North-Holland

Benchmarks for basic scheduling problems

E. T a i l l a r d
Ecole Polytechnique Fdddrale de Lausanne, Ddpartement de Math~matiques, CH-1015 Lausanne,
Switzerland

Received December 1989; revised March 1992

Abstract: In this paper, we propose 260 randomly generated scheduling problems whose size is greater
than that of the rare examples published. Such sizes correspond to real dimensions of industrial
problems. The types of problems that we propose are: the permutation flow shop, the job shop and the
open shop scheduling problems. We restrict ourselves to basic problems: the processing times are fixed,
there are neither set-up times nor due dates nor release dates, etc. Then, the objective is the
minimization of the makespan.

Keywords: Combinatorial optimization; Scheduling; Benchmarks

Introduction sists in finding a permutation of the n jobs that

minimizes the makespan.
The types of problems discussed in this paper In the case of the job shop problem, any pro-
(permutation flow shop, job shop and open shop cessing order of the jobs on the machines is
scheduling problems) have been widely studied in allowed. For every job, the operations must be
the literature using exact or heuristic methods, processed in a given order on the machines, but
but a common comparison base is missing. We this order may differ according to the jobs.
hope that this paper will fill a gap in this domain. For the open shop problem, every operation is
The three-field nomenclature described in assigned to a given machine but the order of the
Lawler et al. [9] names these problems F II Cmax, operations of every job is totally free.
J II Cmaxand O[I Cmax respectively. They certainly The aim of this paper is to present unsolved
belong to the most studied ones among schedul- problems whose size corresponds to the one of
ing problems. Let us describe them briefly. industrial problems. These problems must be easy
There are n jobs that have to be performed on to generate.
m unrelated machines; in our case, every job
consists of m nonpreemptible operations. Every
operation of a job uses a different machine dur-
Generating interesting problems
ing a given time and may wait before being pro-
cessed.
For the permutation flow shop problem, the As we do not know any method to solve ex-
operations of every job must be processed on actly the problems we want to propose, we have
machines 1 , . . . , m in this order. Moreover, the used heuristic methods to get hopefully good
processing order of the jobs on the machines is solutions of these problems. These heuristic
the same for every machine. The problem con- methods are based on taboo search techniques.
Taboo search is described very generally in Glover
[7] and one can find some of its practical applica-
Correspondence to: E. Taillard, l~cole PolytechniqueF6d6rale tions to the flow shop sequencing problem in
de Lausanne, D6partement de Math6matiques, CH-1015 Lau- Taillard [13] and Widmer and Hertz [14], and to
sanne, Switzerland email: TAILLARD@MAS65.EPFL.CH. the job shop scheduling problem in Taillard [12].

0377-2217/93/$06.00 © 1993 - Elsevier Science Publishers B.V. All rights reserved

 E. Taillard / Benchmarksfor basic schedulingproblems 279

Taboo search is very easy to implement and can The random number generator
provides results that are better than those ob-
tained by any other heuristic method described in Let us recall the implementation of the linear
[9] or in Applegate and Cook [1] if its parameters congruential generator we have used which is
are well tuned. based on the recursive formula
In order to propose problems that are as diffi-
cult as possible (the most interesting ones) we Xi+ 1 = (16807Xi) m o d ( 2 3 1 - 1).
have generated many instances of problems that This implementation uses only 32-bit integers and
we have 'solved' in a summary way with taboo provides a uniformly distribution sequence of
search. Then, we have chosen the 10 problems numbers between 0 and 1 (not contained):
that seemed to be the hardest ones and we have Step O. Initial seed and constants:
solved them once more, allowing our heuristic X 0 ( O < X 0 < 2 31-1);
method to perform a higher number of iteration. a = 16807, b = 127773, c = 2836, m = 231 - 1.
Obviously, the choice of the hardest problems Step 1. Modification of the seed:
is very subjective. We decided that a problem was
interesting if the best makespan we found was far k := [ X J b I ,
from a lower bound of the makespans and if
Xi+ 1 : = a ( X i rood b) - k c ,
many attempts to solve the problem (starting
from various initial solutions) did not provide the If X i+1 < 0, then let
same solution. Such a method enabled us to
Xi+l:=Xi+l + m .
detect the simplest problems but we may not
propose problems that have a local optimum with Step 2. New value of the seed: Xi+ 1.
a large attraction basin. Current value of the generator: Xi+ l/m.
Below, we shall denote by U(0, 1) the pseudo-
random number that this generator provides. We
have 0 < U(0, 1) < 1 for every generated number.
The problems We shall denote by U[a, b] (with a < b, a and
b integers) the integer number

The problems we propose are randomly gener- ta + U(O, 1) " ( b - a + 1)].

ated with a good random number generator pro- For every random integer generated, we have
posed in Bratley [4]. We recall its implementation
so that this paper is self contained. a < U[a, b] < b
A problem will be entirely defined by the
and every integer between a and b has the 'same'
initial value of the seed of the random generator probability of being chosen. In order to imple-
and by the way of generating it. For every type of ment the integer random procedure only with
problem, we give a simple manner of computing a 32-bit integers, the problems have been chosen in
lower bound of the makespan; in particular, this such a way that one never has to deal with a seed
permits to verify the generation of the problem. X such that
For every size of problem, we give the total
number of instances we have generated (summary [a+
X ' ( b - a + l --
) l -4: [a+ X 1
resolution), the maximum number of iteration of m [m/(b - a + 1)]
taboo search that were done (long resolution) and
the proportion of problems that were solved up
to the lower bound, that is to say optimally. For
every type and every size of problem, we give 10 Flow shop problems
instances. For each instance, we give the initial
value of the random generator seed, the best There are in the literature some problems of
value of the makespan we have found (i.e. an this type; let us quote for example eight small and
upper bound of the optimal makespan) and a simple problems proposed in Carlier [5] and
lower bound of all the makespans. solved exactly in this reference.
280 E. Taillard / Benchmarks for basic schedulingproblems

The flow shop problems are characterized by work up to the end of the operations, and let T~
the processing times dij. of job j on machine i be its total processing time. We have:
(1 <_ i < m , 1 < j < n). We have generated the val-
i-1 )
ues of dij by the following way:
For i = 1 to m
for j = 1 to n
dij := U[1, 99].
ai=min. ~ dk~ ,
We propose problems with 5, 10 and 20 ma- J k=i+l
chines and from 20 to 500 jobs. We compute the
lower bound of the makespan as presented be-
low. j=l
Let b i be the minimum amount of time before
machine i starts to work and a i be the minimum Clearly, the optimal makespan Cm~
* , is greater
amount of time that it remains inactive after its than or equal to the maximum between the mini-

Time seed UB LB Time seed UB LB Time seed UB

20jobs 5machines 50jobs 10 machines 100jobs 20 machines

873654221 1278 1232 1958948863 3037 2907 450926852 6330 5851
379008056 1359 1290 575633267 2911 2821 1462772409 6320 6099
1866992158 1081 1073 655816003 2871 2801 1021685265 6364 6099
216771124 1293 1268 1977864101 3067 2968 83696007 6331 6072
495070989 1235 1198 93805469 3011 2908 508154254 6405 6009
402959317 1195 1180 1803345551 3021 2941 1861070898 6487 6144
1369363414 1239 1226 49612559 3124 3062 26482542 6379 5991
2021925980 1206 1170 1899802599 3048 2959 444956424 6514 6084
573109518 1 2 3 0 1206 2013025619 2910 2795 2115448041 6386 5979
88325120 1108 1082 578962478 3 1 0 0 3046 118254244 6534 6298
20jobs 10machines 50jobs 20machines 200 jobs 10 machines
587595453 1582 1448 1539989115 3886 3480 471503978 10872 10816
1401007982 1659 1479 691823909 3733 3424 1215892992 10500 10422
873136276 1496 1407 655816003 3673 3351 135346136 10956 10886
268827376 1377 1308 1315102446 3755 3336 1602504050 10893 10794
1634173168 1419 1325 1949668355 3648 3313 160037322 10537 10437
691823909 1397 1290 1923497586 3719 3460 551454346 10347 10255
73807235 1484 1388 1805594913 3 7 3 0 3427 519485142 10882 10761
1273398721 1538 1363 1861070898 3737 3383 383947510 10754 10663
2065119309 1593 1472 715643788 3772 3457 1968171878 10465 10348
1672900551 1591 1356 464843328 3791 3438 540872513 10727 10616
20jobs 20machines 100jobs 5 machines 200 jobs 20 machines
479340445 2 2 9 7 1911 896678084 5493 5437 2013025619 11393 10979
268827376 2099 1711 1179439976 5274 5208 475051709 11445 10947
1958948863 2326 1844 1122278347 5175 5130 914834335 11522 11150
918272953 2 2 2 3 1810 416756875 5018 4963 810642687 11461 11127
555010963 2291 1899 267829958 5250 5195 1019331795 11427 11132
2010851491 2226 1875 1835213917 5135 5063 2056065863 11368 11085
1519833303 2 2 7 3 1875! 1328833962 5247 5198 1342855162 11536 11194
1748670931 2 2 0 0 1880 1418570761 5 0 9 4 5038 1325809384 11544 11126
1923497586 2237 1840 161033112 5 4 4 8 5385 1988803007 11424 10965
1829909967 2 1 7 8 1900 304212574 5328 5272 765656702 11548 11122
50jobs 5machines 100jobs I0 machines 500jobs 20 machines
1328042058 2724 2712 1539989115 5776 5759 1368624604 26316 25922
200382020 2 8 3 6 2808 655816003 5362 5345 450181436 26807 26353
496319842 2621 2596 960914243 5679 5623 1927888393 26626 26320
1203030903 2751 2740 1915696806 5 8 2 0 5732 1759567256 26642 26424
1730708564 2863 2837 2013025619 5491 5431 606425239 26509 26181
450926852 2 8 2 9 2793 1168140026 5308 5246 19268348 26654 26401
1303135678 2725 2689 1923497586 5600 5523 1298201670 26575 26300
1273398721 2 6 8 3 2667 167698528 5640 5556 2041736264 26794 26429
587288402 2 5 5 4 2527 1528387973 5891 5779 379756761 26241 25891
248421594 2 7 8 2 2776 993794175 5856 5830 28837162 26662 26315

Figure 1. Instances of flowshop problems

 E. Taillard / Benchmarks for basic scheduling problems 281

mum amount of time required by the machines Table 1

Flow shop problems
and the minimum of time required f o r each job:
Nb Nb Nb LB Nb Nb
jobs ma- in- reached itera- resolu-
LB = max m y ( b i + T i + ai), max dij chines stances (%) tions tions
20 5 100 35 104 3
<_ Cma x . 20 10 100 1 104 3
This lower bound is easy to compute and we 20 20 100 0 2-104 3
50 5 70 41 5.103 3
conjecture that for fixed m 50 10 70 3 104 3
50 20 70 0 5-104 3
lim Prob(Cm*ax = LB) = 1.
n ----~ 0 o 100 5 10000 54 2.103 4
100 10 50 6 2.104 3
For every size of problem we give the following 100 20 50 0 104 3
information (Table 1): 200 10 300 28 4103 3
Nb jobs: The number of jobs. 200 20 25 0 4.103 10
Nb machines: The number of machines. 500 20 100 14 a 4-103 5
Nb instances: The total number of problems a T h e value reached for this size was less than or equal to
generated. 1.02 times the lower bound.
LB reached: The proportion of problems for
which we found a solution for Thompson [11]; most of the optimal values of
which the makespan was equal to these problems are given in [1] or Carlier and
the lower bound (or equal to the Pinson [6]. Some very good solution values of
lower bound augmented by 2% Lawrence's problems are also given in [12]. We
for the 500-job 20-machine prob- can consider that problems up to ten machines
lems). may be solved satisfactorily with existing meth-
Nb iterations: The maximum number of itera- ods. This is why we propose problems with 15 and
tions performed by taboo search 20 machines and from 15 to 100 jobs.
(long resolution). The processing time dij of the i-th operation
Nb resolutions: The number of attempts to solve of job j (1 _<i < m , 1 < j < n ) is obtained as fol-
the problem from various initial lows:
solutions (long resolution). For j = 1 to n
Then we give ten instances for every size of For i = 1 to m
problem with the following information (Figure dii := U[1, 99].
1): The machine Mii on which the i-th operation
Time seed: The initial value of the random gener- of job j has to be performed is given by the
ator's seed. following procedure:
UB: An upper bound of the optimal makespan Step O.
(the best value we got). Mij :=i (1 < i < m , 1 < j < n ) .
LB: A lower bound of the makespans.
As the aim is to give an upper bound as good
Table 2
as possible but not a fast solving method, the Job shop problems
computation time does not have much impor-
Nb Nb Nb Lb Nb Nb
tance. However, let us mention that an iteration
jobs ma- in- reached itera- resolu-
of taboo search needs about 4 . 1 0 - 6 n 2 m seconds chines stances (%) tions tions
on a 'Silicon Graphics' personal workstation (10
15 15 50 0 7.105 4
Mips). 20 15 50 4 106 3
20 20 50 0 107 4
30 15 50 26 3.106 4
Job shop problems 30 20 50 0 2.106 4
50 15 100 78 3-106 4
In the literature, we may find instances of 50 20 26 27 5-105 4
100 20 100 97 5-105 3
small problems in Lawrence [10] and Muth and
282 E. Taillard / Benchmarks for basic scheduling problems

Step 1.
For j = 1 to n
For i = 1 to m a i=min.
{m)
with k ' such that Mk, j = i;

~ dk~ ,
Swap Mij and Mv[i.ml j. J k=k'+l
Let us note the use of another initial seed for
the choice of the machines: machine seed. with k ' such that Mk, j = i;
An instance of a small open shop problem,
obtained with the same procedures, is given ex- T, = ~ dk,j,
tensively in Table 4. ]=1
The lower bound is computed as previously with k ' such that Mk, j = i.
but bi, a i and T/ are defined as follows: We conjecture again that this bound is tight if
k'-I / n / m ~ 0% because we have always found an opti-
bi=m) n k~_ldk]} , mal schedule if n / m > 6, considering more than
2000 problems whose size was varying from 20

Time seed Machine seed UB LB Time seed Machine seed UB LB

15jobs 15machines 30jobs 20machines

840612802 398197754 1231" 1005 1841414609 1357882888 2 0 6 4 1850
1314640371 386720536 1252 953 2116959593 1546338557 1 9 8 3 1761
1227221349 316176388 1223 1036 796392706 1230864158 1 8 9 6 1710
342269428 1806358582 1181 973 532496463 254174057 2 0 3 1 1820
1603221416 1501949241 1234 940 2020525633 978943053 2 0 3 2 1785
1357584978 1734077082 1243 1134 52~444252 185526083 2 0 5 7 1940
44531661 1374316395 1228 1103 1569394691 487269855 1 9 4 7 1751
302545136 2092186050 1221 980 1460267840 1631446539 2 0 0 5 1770
1153780144 1393392374 1289 1020 198324822 1937476577 2 0 1 3 1758
73896786 1544979948 1261 940 38071822 1541985579 1 9 7 3 1678
20jobs 15 machines 50jobs 15 machines
533484900 317419073 1376 1254 17271 718939 2760* 2760
1894307698 1474268163 1381 1267 660481279 449650254 2756* 2756
874340513 509669280 1367 1243 352229765 949737911 2717" 2717
1124986343 1209573668 1355 1329 1197518780 166840558 2839* 2813
1463788335 529048107 1366 1163 1376020303 483922052 2 6 8 9 2679
1056908795 25321885 1371 1211 2106639239 955932362 2781~ 2781
195672285 1717580117 1480 1306 1765352082 1209982549 2943, 2943
961965583 1353003786 1432 1315 1105092880 1349003108 2885, 2885
1610169733 1734469503 1361 1202 907248070 919544535 2655, 2655
532794656 998486810 1373 1213 2011630757 1845447001 2723 2723
20jobs 20machines 50jobs 20machines
1035939303 773961798 1663 1217 8493988 2738939 2921 2868
5997802 1872541150 1626 1314 1991925010 709517751 3 0 0 2 2848
1357503601 722225039 1574 1248 342093237 786960785 2 8 3 5 2755
806159563 1166962073 1660 1284 1634043183 973178279 2 7 7 5 2697
1902815253 1879990068 1598 1256 341706507 286513148 2 8 0 0 2725
1503184031 1850351876 1679 1245 320167954 1411193018 2 9 1 4 2845
1032645967 99711329 1704 1403 1089696753 298068750 2 8 9 5 2812
229894219 1158117804 1626 1387 433032965 1589656152 2 8 3 5 2764
823349822 108033225 1635 1352 615974477 331205412 3 0 9 7 3071
1297900341 489486403 1614 1277 236150141 592292984 3 0 7 5 2995
30jobs 15 machines 100jobs 20machines
98640593 1981283465 1770 1764 302034063 1203569070 5464* 5464
1839268120 248890888 1853 1774 1437643198 1692025209 5181" 5181
573875290 2081512253 1855 1733 1792475497 1039908559 5568* 5552
1670898570 788294565 1851 1828 1647273132 1012841433 5339* 5339
1118914567 1074349202 2007 1754 696480901 1689682358 5392* 5392
178750207 294279708 1844 1777 1785569423 1092647459 5342* 5342
1549372605 596993084 1822 1771 117806902 739059626 5436* 5436
798174738 151685779 1714 1673 1639154709 1319962509 5394* 5394
553410952 1329272528 1824 1764 2007423389 749368241 5358* 5358
1661531649 1173386294 1723 1608 682761130 262763021 5 2 1 3 5183

Figure 2. Instances of job shop problems

 E. Taillard / Benchmarks for basic scheduling problems 283

Time seed Machine seed UB LB Time seed Machine seed UB LB

4jobs 4 machines 10jobs 10 machines

1166510396 164000672 193" 186 1344106948 1868311537 645 b 637
1624514147 1076870026 236* 229 425990073 1111853152 588 *b 588
1116611914 1729673136 271" 262 666128954 1750328066 611 b 598
410579806 1453014524 250* 245 442723456 1369177184 577 *b 577
1036100146 375655500 295* 287 2033800800 1344077538 641 b 640
597897640 322140729 189" 185 964467313 1735817385 538 *b 538
1268670769 556009645 201" 197 1004528509 967002400 623 616
307928077 421384574 217" 212 1667495107 818777384 596 b 595
667545295 485515899 261" 258 1806968543 1561913259 595 *b 595
35780816 492238933 217" 213 938376228 344628625 602b 596
5jobs 5machines 15jobs 15 machines
527556884 1343124817 300* 295 1561423441 1787167f~67 937 *b 937
1046824493 1973406531 262* 255 204120997 213027331 918 *b 918
1165033492 86711717 323 *a 321 801158374 1812110433 871 *b 871
476292817 24463110 310" 306 1502847623 1527847153 934 *b 934
1181363416 606981348 326% 321 282791231 1855451778 950b 946
897739730 513119113 312" 307 1130361878 849417380 933 *b 933
577107303 2046387124 303*a 298 379464508 944419714 891 *b 891
1714191910 1928475945 300* 292 1760142791 1955448160 893*b 893
1813128617 2091141708 353* 349 1993140927 179408412 908 b 899
808919936 183753764 326* 321 1678386613 1567160817 902*b 902
7jobs 7 machines 20jobs 20 machines
1840686215 1827454623 438 435 957638 9237185 1155*b 1155
1026771938 1312166461 449 443 162587311 1489531109 1244b 1241
609471574 670843185 479 468 965299017 1054695706 1257*b 1257
1022295947 398226875 467 463 1158457671 1499999517 1248*b 1248
1513073047 1250759651 419 416 1191143707 1530757746 1256*b 1256
1612211197 95606345 460 451 1826671743 901609771 1209b 1204
435024109 1118234860 435 422 1591533998 1146547719 1294*b 1294
1760865440 1099909092 426 424 937297777 92726463 1173b 1169
122574075 10979313 460 458 687896268 1731298717 1289*b 1289
248031774 1685251301 400 398 687034842 684013066 1241*b 1241

Figure 3. Instances of open shop problems

jobs, 2 machines to 150 jobs, 15 machines, passing chine seed in Figure 2. Stars in Figure 2 indicate
by 500 jobs, 4 machines. optimum values.
The time needed to perform one iteration of
taboo search is about 20.10-6nm seconds on
Open shop problems
'Silicon Graphics' workstation. Table 2 and Fig-
ure 2 are analogous to Table 1 and Figure 1, but, We do not know instances of such problems in
for job shop problems, we give in addition m a - the literature. This is why we give problems of

Machine

El ~~',/'7,~,)fS~,,TY'///'~Y/";///"
Y,;//,V,, 7/';. 7.:~,)S'Z[ 2 : 3 ; :): ;:;:: ,',:t

Irl Ba
]
I I I

50 100 150 Time

Figure 4. An optimum schedule for the first problem of Table 4

284 E. Taillard/ Benchmarksfor basicschedulingproblems
small size. These problems are obtained using which the number of jobs is about the number of
exactly the same procedures as those used for the machines. In this case, either an optimal solution
job shop problems, and the lower bound of the is easily reached, or the problem is harder than
makespans corresponds to the maximum amount job shop problems of the same size.
of time that a job or a machine requires, i.e. For problems with n >> m, we have observed
empirically that the mean complexity of taboo
search applied to open shop problems, O(n 2"37.
LB=max(nliax(~dij)' rtllax(i,klMEik=jdik ) m369), is lower than the complexity of taboo
search applied to job shop problems, O(n 2"5°.
Because one has to choose the order of the m3.81).
operations of a job, one can find very often an In Table 4, we describe extensively the first
optimal schedule, except for the problems in 4-job 4-machine problem we propose, i.e. the

function unif(var seed: integer; low, high: Integer): Integer;

const
m = 2147483647; a = 16807;
b = 127773; c = 2836;
var
k: Integer;
value_ 0_1: double; (* floating point coded on 64 bits *)
begin
k := seed dlv b;
seed := a * (seed m o d b) - k * c;
If seed < 0 then seed := s e e d + m ;
value_ 0_1 := s e e d / m ;
unif := low + trunc(value_ O_ 1 • (high - low + 1))
end;
procedure generate_flow_ shop( var time_ seed: Integer;
nb_ jobs, nb_ machines: Integer;
v a t d: matrix);
(* type matrix =array[l..20, 1-500] of Integer; must be declared above *)
var i, j: Integer;
begin
for i := t to nb_ machines do
for j : = I to n b _ j o b s do
d[i, j] := unif(time_ seed, I, 99)
end;
procedure generate_ job_ and _ open_ shop( var time_ seed, machine_ seed: Integer;
nb_ jobs, nb_ machines: Integer;
var d, M: matrix);
var i, j: Integer;
procedure swap(var a, b: Integer);
var temp: Integer;
begin
temp := a; a := b; b := temp
end;
begin
for j ,= 1 to nb_ jobs do
for i := 1 to nb_ machines do
d[i, j] := unif(time_ seed, 1, 99);
for j := 1 to nb_ jobs do
for i := 1 to nb_ machines do
M[i, j] : - i;
for j := 1 to nb_ jobs do
for i := 1 to nb_ machines do
swap(M[/, j], M[unif(machine_ seed, i, nb_ machines), j])
end;

Figure 5. Pascal code for the generation of scheduling problems

 E. Taillard / Benchmarks for basic scheduling problems 285

Table 3 are proved optimal or not, in order to update the

Open shop problems best solutions known. The data included in Fig-
Nb Nb Nb LB Nb Nb ures 1 to 3 are available via e-mail in the OR-
jobs ma- in- reached itera- resolu- Library created by Beasley [2].
chines stances (%) tions tions
4 4 50000 98.5 105 5
5 5 45000 99.7 5-10 s 4
7 7 1000 94 106 5 References
10 10 300 89 2.106 5
15 15 40 52 3.105 3
20 20 25 24 3"106 3 [1] Applegate, D., and Cook, W., "A computational study of
the job-shop scheduling problem", ORSA Journal on
Computing 3 (1991) 149-156.
[2] Beasley, J.E., "OR-Library: Distributing test problems by
processing times dij of the operation i of job j electronic mail", Journal of the Operational Research
and its associated machine Mij. We give an opti- Society 41 (1990) 1069-1072.
mum schedule of this problem in the Gantt chart [3] Br~isel, H., Tautenhahn, T., and Werner, F., "Construc-
of Figure 4. tive heuristic algorithms for the open shop problem",
Math 21/91, Institut ffir Mathematische Optimierung,
The time needed by taboo search to perform Technische Universit~it 'Otto yon Guericke', Magdeburg,
one iteration is about 23.10-6nm seconds. Ta- 1991.
bles 3 and Figure 3 are analogous to Table 2 and [4] Bratley, P., Fox, B.L., and Schrage, L.E., " A guide to
Figure 2. Stars in Figure 3 indicate optimum Simulation", Springer-Verlag, New York, 1983.
values; a) indicates solution values issued from [5] Carlier, J., "Probl~mes d'ordonnancement ~t contraintes
de ressources: Algorithmes et complexitE', M&hodologie
Kleinau [8] who proves optimality for 4 x 4 and
et architecture des syst~mes informatiques, Institut de
5 x 5 open shop problems; b) indicates solution programmation, Universit6 P.&M. Curie, Paris, 1984.
values issued from Br~isel et al. [3]. [6] Carlier, J., and Pinson, E., "An algorithm for solving the
Finally, we give in Figure 5 the procedures, job-shop problem", Management Science 35/2 (1989)
written in Pascal language, for generating the 164-176.
[7] Glover, F., "Tabu Search - Part I", ORSA Journal on
scheduling problems described in this paper. Computing 1 (1989) 190-206.
[8] Kleinau, U., "On some new methods for solving machine
scheduling problems", Math 16/91, Institut fiJr Mathe-
Concluding remarks matische Optimierung, Technische Universit~it 'Otto von
Guericke', Magdeburg, 1991.
[9] Lawler, E.L., Lenstra, J.K., Rinnooy Kan, A.H.G., and
We hope that the problems that we propose Shmoys, D.B., "Sequencing and scheduling", Report BS-
will constitute a comparison base for future reso- R89xx, Centre for Mathematics and Computer Science,
lution methods. Amsterdam, Netherlands, 1989.
Everyone may send us his own results about [10] Lawrence, S., "Resource constrained project scheduling:
these problems, specifying whether his solutions An experimental investigation of heuristic scheduling
techniques (supplement)", Graduate School of Industrial
Administration, Carnegie Mellon University, 1982.
Table 4 [11] Muth, J.F., and Thompson, G.L., "Industrial scheduling",
The first instance of the 4-job 4-machine open shop problem Prentice-Hall, Englewood Cliffs, NJ 1963, 225-251.
[12] Taillard, E., "Parallel taboo search technique for the job
Opera- Job j a Job j b shop scheduling problem", ORWP 89/11, DMA, l~cole
tion i 1 2 3 4 1 2 3 4 Polytechnique F~d~rale de Lausanne, Lausanne, Switzer-
land, 1989.
1 54 9 38 95 3 4 1 1
[13] Taillard, E., "Some efficient heuristic methods for the
2 34 15 19 34 1 1 2 3
flowshop sequencing problem", European Journal of Op-
3 61 89 28 7 4 2 3 2
erational Research 47 (1990) 65-74.
4 2 70 87 29 2 3 4 4
[14] Widmer, M., and Hertz, A., "A new heuristic for the
a Processing times dip flowshop sequencing problem", European Journal of Op-
b Machines Mi~. erational Research 41 (1989) 186-193.