Sie sind auf Seite 1von 9

A revised proof of optima&y for the cube-per-order index rule for stored item location

Charles J. Malmborg and Krishnakumar Bhaskaran


Department of Decision Sciences and Engineering Systems, Institute, Troy, NY, USA
The cube-per-order

Rensselaer

Polytechnic

index (COl) is a commonly used sturuge assignment policy for u//ocating inventory The optimality of COI-based assignment policy *for a dual-address order-picking discipline (with storage and rethcval interleuving) has been established. Additionul empirical evidence based on recent studies and counterexamples reveul the nonuniqueness of the COI layout and questions the validity of the earlier proof. A completely revised proof of optimality of the CO1 assignment policy for dual-address systems is presented, taking into account the nonuniqueness sf the CO1 layout. A heuristic is suggested to identifj, the optimal COI layout from the set of COI layouts. Empirical evidence of the optimality of the CO1 layout in comparison with the fLlmi(ior L-shaped and Organ Pipe configurations is provided.
in a warehouse. Keywords: warehousing. layout. cube per order index, order
picking, optimization

Introduction
The cube-per-order index (COl) is a very widely used rule of thumb for allocating storage space to inventoried items in a warehouse. It is the ratio of the items storage space requirement (cube) to its popularity (number of storage/retrieval requests for the item). The COI-based assignment policy ranks the items on the basis of their CO1 values, in an ascending order, and then allocates them in that order to the most accessible locations, that is, locations closest to the input/output (l/O) point.le3 Harmatuck4 has shown that the COI-based item allocation minimizes the expected order-picking cost for at least one linear programming formulation of the single-address stock location models.s-7 More recently, Malmborg and BhaskaranX have demonstrated the order-picking cost optimality of CO1 when vehicles are routed to serve three or more transactions during the picking cycle. The optimality of the COI-based assignment policy for the case in which a single storage transaction and a single retrieval transaction are interleaved in order to economize on the travel of the order-picking vehicle has also been investigated by

Malmborg and Bhaskaran;9 their formulation dual-address model assumes:


I. unconstrained

of the

2.

3. 4. 5.

rectilinear travel of the order-picking vehicle, that one storage transaction and one retrieval transaction are paired together on each order-picking cycle, that is, the system uses dual-address cycles, statistical independence between the storage and retrieval transactions that are interleaved, that the cost of order-picking is proportiona to the distance travelled by the order-picking vehicle, and that there are no incompatibility constraints in locating the items.

The dual-address cycle defined in Ref. 9 for the warehousing situation shown in Figure I encompasses the travel of the order-picking vehicle to obtain a unit load from the storage queue, travelling to the appropriate storage address, depositing the unit load, travelling directly to the retrieve location, picking the unit load at the retrieve location, and transporting it back to I/O point, where it is deposited on a conveyor. The expected order-picking cost for such a dual-address system, based on the notations listed below, was defined in Ref. 9 as Z =
f 2 bj fjW,iQj) ( lo/, + c afh (L + LX-, > (1)

Address Decision Institute, Received

reprint requests to Dr. Malmborg at the Department of Sciences and Engineering Systems, Rensselaer Polytechnic Troy, New York 12180-3590, USA. 27 April 1989; accepted 27 September 1989

where b = location index, j=itemsindex(j=

l,...,N),

1990 Butterworth

Publishers

Appl. Math. Modelling,

1990, Vol. 14, February

87

Revised proof of optimality

of COl assignment
AREA

policy: C. J. Malmborg

and K. Bhaskaran

Theorem 1
RESERVE STORAGE

This theorem establishes the optimality of the COIbased assignment policy for the dual-address system alluded to earlier.
STAGING . . AREA . . .

Stutement

The CO1 storage assignment rule will produce a least cost solution to the cost equation given by (1) for fixed inventory levels.
M

Racks

numbe,ed

1 lhrough

Proof

of Theorem

I
OUT 110 Porn

Since the order-picking cost is proportional to the distance travelled by the order-picking vehicle, t may be eliminated from expression (1) to yield the following equation: z C.f~X/JQ~) lob + C (Iho + Im,)Pu ( nfh > In the above, the assumption of fixed inventory makes it possible to express Qj as Qj = XX,, = _fjSjdj (3) levels (4) size for itemj and

FRONT VIEW OF RACK (i.e. RACK FACE)

Figure 1.

Warehousing

situation

where Sj = the average transaction cost per unit distance of travel of the order-picking vehicle, Lh = rectilinear distance from l/O to location h, lhr, = rectilinear distance from b + a to LI, rectilinear distance from a to I/O, Lo = average frequency of transactions for fj = item j, Xhj = quantity of item j located in location b, Qj = lot size of item j, P, = normalization frequency of access of location ~1.
t=

dj = the number of periods demand stored for itemj.

Substituting for Q, in (3), simplifying, and augmenting with location capacity and nonnegativity constraints yield the following quadratic assignment problem: Minimize: Z = zx
h .i

(Xbil(sjdj))
(

Ioh

x o#h

(Itxr

~cu,)Po 1 (.v

subject to CCUj*Xhj(
XbjlO

Mh

for all b

(6)

In the above, P,, is defined as the ratio of the number of transactions per period from the items stored in location ~1to the total number of transactions for all inventoried items in a period, that is,

for all bj

(7)

The optimality of the COl-based assignment policy for the dual-address system described above is presented in Ref. 9; the theorem and proof outlined there implicitly assume a unique CO1 layout. Recent empirical studies conducted by BhaskaranO and the counterexamples by Bozer indicate the nonuniqueness of the CO1 layout, thereby questioning the validity of the proof of optimality of CO1 for dual-address systems. The following sections present a completely refined theorem and proof that take into consideration the recent empirical findings. Additionally, heuristics are suggested to pinpoint the optimal CO1 layout. The optimality of the CO1 layout is then empirically demonstrated in comparison with the familiar L-shaped and Organ Pipe configurations. The final section summarizes the findings.

where Mh = capacity of storage location b. The above minimization problem has constraints similar to the linear programming formulation for simple out-and-back storage reported in Refs. 4 and 5, but differs with respect to the objective function due to the second term &+,(lha + /,,)P,. The first term of the objective function Chj(Xbj/Sjdj)loh represents a simple out-and-back order-picking cost for which the optimality of CO1 is demonstrated in Ref. 4. The proof of optimality of CO1 for the dual-address case is established by using the following reasoning: policy (that is, fixed xs) the assignment of the highest P values to the locations closest to the I/O point is the optimal policy for the dual address case, and 2. CO1 is one such policy. By invoking again the fixed inventory level assumption, the P value, that is, the normalized frequency of
1. for a given storage assignment

88

Appl.

Math.

Modelling,

1990, Vol. 14, February

Revised

proof

of optimality

of COl assignment

policy:

C. J. Malmborg

and K. Bhaskaran

access of a location,
j

is redefined as
j (8)

P, = C (Xyj/sjdjD) = C (XjCUjl[(COIj)(D)]

where COIj = (cUjQ,f,) = (CUjsjdj>and D = ZIj.fj. For a given assignment policy such as the COI, the Xs are fixed, and hence the P-values are inversely related to the COI. A location containing items with lower CO1 values will therefore have a higher P-value. The CO1 assignment policy assigns items with lowest CO1 closest to the I/O point. Hence the CO1 policy assigns the highest P-values to locations closest to the I/O point. This establishes the latter half of the reasoning. Consider next the following proposition.
Proposition

Figure 2.

Tours affected by swapping

Consider any two locations e and f such that I,,,, < l,f, that is, e is closer to the I/O point than f. If a unit volume of item is swapped between the two locations in violation of the COI, then this results in increasing the summation of the product &,.L,(L, + UP,,.
Proof

VO

Figure 3.

q irrl
n

,,

II

If

II

.,

,*

VO

Properties of rectilinear travel

By definition,
P, = 2 (X<~cUjl[(COIj)(D)]

(9)

.I where (X,jcq) is the volume of item j in location P. According to the CO1 policy, if l,,, < f,,/, then coI,j 5 COIL foralljEr, kEf (10) If a unit volume of itemj in location e is swapped with a unit volume of item k in location f, then the change in the P-value of the location e is
pnew = pa

The set of tours exhaustively covers all the tours that are affected by the swapping. The tours are sketched in Figure 2. Consider the pair of tours between locations e and f. The change in the product for the tour o + e -+ f-tois

U,..f + LfWY

- Pid) = + (Lf

l,,)IPp

- PYd(

(14) and the change in the product for the complementary tour 0 --f f -+ e -i; 0 is (If<> + I,,,>)(P=- P$d) = -(If,
+

<

[(CO!,)(D)1 3- [KOI,W)l

- (11)

that is,
pold c _ Ppw = [(l/COIj)

I,,,)jPy

- (IlCOI~)J(llD)

2 0 (12)

- P:I\ (15)

This implies that the P-value of the closer location decreases. Similarly, the corresponding change in the P-value of the more distant location f is PN _ fmld = [(l/CO!,(llC01,)](1/D)~O
.f

Therefore the net change in the product considering the tours between locations e and f is [l,,f +
l,,f If, -

/,,,.11P? - Pyq

(16)

since the absolute change in the P-values is the same. The above simplifies to (I,,, - f,,,,)JP) - Pqdl (17) Since l,,f > I,,, by assumption, the above change is positive; that is, an increase in the product. Before studying the impact of swapping on other pairs of tours, certain properties of the rectilinear travel are developed. Consider any two locations m and n as shown in Figure 3. If I,, < I,,,,, then = 21*, - l,, L, + L = (6 - 10,) + l<,,n
(18)

(13) This implies that the P-value of the more distant location f increases by an amount identical to the decrease in P-value of location e. Therefore the absolute change in P-values at each location is the same. To study the impact of swapping unit volume of items on the summation of the product, the following tours need to be considered: 1. the complementary pair of tours o 4 f + e * o and o + e + f 3 o, where o is the I/O point, and 2. the remaining complementary pairs of tours o + c+e~oando-+c+f-+o,wherec#eorf and includes all other locations.

If I,,,, > I,,, then i,,,, + I,, = lo,. To cover the other pairs of tours exhaustively, three mutually exhaustive cases are investigated as shown in Figure 4: Case I: l,, < l,< < l,,f (19)

Appl.

Math.

Modelling,

1990, Vol.

14,

February

89

Revised proof of optimality

of CO/ assignment

policy: C. J. Malmborg

and K. Bhaskaran

to the I/O point is the optimal policy for the dualaddress case. It has already been shown that CO1 policy assigns the highest P-values to locations closest to the I/O point. The proof of optimality of CO1 for the dual-address case is therefore complete.

Empirical insights into the optimality

of CO1

Figure 4.

Other mutually

exhaustive

complementary

tours

Case II: l,, > I,,. < lof and Case III: I,, < l,, > l,,

(20)

The optimality of the cube-per-order index (COI)-based assignment policy for dual-address work cycles was established in Theorem 1. The objectives of this section are to provide additional empirical evidence of the optimality and to highlight the fresh insights gained from the empirical study. The contrived examples used in the study are designated by problem numbers. The notations used are consistent with that used in Theorem 1. The discussion follows the sequence of problems that are empirically analyzed.
Problem 1

(21)

The characteristics

of the problem are

where c # e or f and includes all other locations. Each case is investigated individually.
Case Z

If I,, > I,,, then l,., + l,,f,,., and 10c< l,,f is equivalent to 1,f + lOf = 21,f - b.. Hence the net change in the product for the complementary pair of tours o * c -+ e+oando+c+f+ois + [2l,f - 10C- f<,,.J/Py - Ppdl
(22)

three item types; unit cube, that is, CU, = I for j = 1, 2, 3; equal access frequencies, l/3 for j = 1, 2, 3; and storage zone represented by a 3 x 5 grid (3 rows and 5 columns) in which each grid element is of unit capacity.
fj =

Since l,f > l,, by assumption, the above change is positive. Hence the impact of swapping on Case I is to increase the product.
Case ZZ

If l,,. < I,,, then l,., + l,, = 21,,, - i,,., and I,,. < l,f is equivalent to l,, + 10f = 210f - l,,.. The net change in the product for the implementation pair of tours is + [21,, - I,, - 21,, i l,,,] jZyw - Zqdl
(23)

Since lof > l,, by assumption, the impact of swapping on Case II is to increase the product.
Case ZZZ

If L > loo then I,, + l,, = l,,., and i,,. > [O,fis equivalent to lcf + lOf = l,,.. Hence the net change in the product for the complementary pair of tours is
+ [l,, - l,,]JPy - Pqq = 0
(24)

The problem characteristics are shown in Figure 5, and the layout of storage addresses is sketched in Figure 6. The rectilinear distances of the grid elements from the I/O point are recorded within the respective element squares. Recall that the COI-based assignment policy suggests the assignment of items with the lowest CO1 to locations closest to the I/O point. On the basis of this policy the assignment of the fourth unit of item 1 is confronted with three possible locations-( 1, I), (2, 2), and (3, I)-giving rise to three COI-based layouts. An important observation to be noted as a consequence of this is the nonuniqueness of the CO1 layout, that is, the COI-based assignment policy yields more than one layout. The three CO1 layouts and their expected dualcommand order-picking costs are exhibited in Figure 7. The costs are different because the layouts are different with respect to the dual-address order-picking discipline. Though there is a strong temptation to dismiss the CO1 policy as nonoptimal at this stage, consider the following hypothesis:
When the CO1 policy yields more than one layout, it contains the optimal layout in the set of layouts. In the event that CO1 yields a unique layout, it is the optimal layout.

This implies that there is no change in the product by swapping for Case III. The overall impact of swapping a unit volume between any two locations e and f on the summation of the product Xazb (lb0 + /,,)I, is to increase this value. This completes the proof of the proposition. Now it has been established, by virtue of the proposition, that for a given storage assignment policy the assignment of the highest P-values to locations closest

Figure 8 displays an L-shaped layout, which is different from the CO1 layout. The results indicate the nonoptimality of this layout in comparison to the CO1 optimal layout (Layout II in Figure 7). This exercise raises an important question: How can the optimal layout be identified from the set of CO1 layouts? The

90

Appl.

Math. Modelling,

1990, Vol. 14, February

Revised proof of optimal@


Item no.

of COl assignment

policy: C. J. Malmborg

and K. Bhaskaran . 3 3

No of locations 4 5 6 of Problem 1

l/f 3 3 3

co1

2
1 2 3 Figure 5. Characteristics 12 15 16

3 \

c
3 4 5 6 7

Cast = 9.7495061727E+OO
L-shaped configuration for Problem 1

Figure 8.

Item No.

No. of locations

l/f

COI

1
1 2 3 4 5

3 3 3
of Problem 2

12 36 60 I

2
Figure 6. Layout and distance array for Problem 1

12 20
Characteristics

3
Figure 9.

UY0IJ-r

following heuristic is suggested in answer to the question:


l

3 \
l

/
\ Cost = 9.7638271604E+OO

Assign items based on CO1 where distances of the grid elements are computed on the basis of the rectilinear metric. To break the tie between candidate locations, compute their euclidean distances to the I/O point.

LAYOUT

II

cost=

9.7338271604E+OO

The basis for this strategy is explained in a later section. At this point it suffices to note that the euclidean distances supplement the information obtained from the rectilinear metric in implementing the COI-based assignment. In Problem I the locations representing a tie are (I, 3), (2, 2), and (3 1). Their euclidean distances to the I/O point are V&, a, and a, respectively. By using the heuristic mentioned above, location (2, 2) is picked for assignment of the fourth unit of item I, yielding Layout II, which in fact is the optimal layout. To investigate the hypothesis and the heuristic further, we consider another problem.
Problem 2

LAYOUT

IPP

cost-

9_7338271604E+OO

Figure 7. COI layouts and cost values for Problem 1. Note that costs of Layouts II and III are equal, since the picking cost of space (2,2) equals the picking cost of space (1, 3)

Problem 2 considers three items with equal access frequencies, as in Problem I. The items are located in a 6 x 6 square grid. The item characteristics are shown in Figure 9, while the layout of storage addresses is sketched in Figure IO. The COI-based assignment of the items yields two tie situations. The first tie is similar to that observed in Problem 1 and is resolved by using the earlier heuristic; the partial assignment is presented in Figure If. Eleven of the twelve units of item 2 are located. The twelfth unit can be placed in any of the

Appl.

Math.

Modelling,

1990,

Vol. 14, February

91

Revised proof of optimality

of COJ assignment

policy: C, J. Malmborg
3 2 2 2 112 112 Cost. 33 33 23 22 33 33 33 cl 23 23 22

and K. Bhaskaran
3 3 3' 3 3 3 \ 2 1 1

22
12 12

33 23 22

3 3 3

, I

1.3007407407E+O1

cost - 1.3007407407E+01

123
Figure 10.

456

3 2

33 33 23 22

33 33 33 33 22 22 0

3 3 3 3 3 3

3 2 2 2 112 1

33

33

Layout and distance array for Problem 2


2 23 22 33 33 23 12 22 3 3 3 3

LAYQUT

NO.

2 112

1
VI
1v

112

Cost -1.3018765432&O

Cast-1.3018765432E+01

II
3 33 33 23 33 33 33 33 23 3 3 3 3 3

I
2

fg3
2 2 2 33 23 22 12 33 33 33 23 3 3 3 3

1 $

.I

III V

2 2 1

22
12

Figure 11.

Partial layout assignment

for Problem 2
CQS? - ?.3046419753E+Ol

six candidate locations along the diagonal of the grid, giving rise to six CO1 layouts. Using the tie-breaking strategy yields two favorable locations, identified by X; these are at equal euclidean distance from the I/O point. The expected dual-command order-picking cost for these two layouts and the remaining four layouts is listed in Figure 12. The two X layouts have identical cost values and are more favorable than the other four layouts. The tie-breaking scheme is therefore validated by Problems I and 2. Figure 13 presents another layout that is different from the CO1 layout, called the Organ Pipe layout, the background of which is described in the appendix. The results for this layout are found to be worse than those of the CO1 optimal layout. The L-shaped layout shown in Figure 14 also performs more poorly than the CO1 optimal. So far, the issue of multiple layouts yielded by CO1 policy has been investigated. Finally, the last part of the hypothesis is considered; in the event that CO1 yields a unique layout, it is the optimal layout. Problem 3 typifies this situation.

Figure 12.

COI layouts and cost values for Problem 2

33

33

-2
21

2
12

2 c 22 22

Figure 13.

Organ Pipe configuration

for Problem 2

92

Appl.

Math.

Modelling,

1990, Vol. 14, February

Revised proof of optimality 3 3 2 t2 33 33 22 22 3 3 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3

of COl assignment

policy: C. J. Malmborg

and K. Bhaskaran

2 3 3 3
1 /

3 -\ 2 2

1 2/ 2 / 3 3 3
\

D
1 1
112 \s
Figure 14.

1costFigure 18.

9.1880364087E+OO

- 13300246913E+01
for Problem 2

Organ Pipe configuration

for Problem 3

L-shaped configuration

Item no. I

No. (

I
Figure 15. Characteristics of Problem 3

271

Problem 3 Problem 3 has layout characteristics identical to those in Problem 1. The item characteristics of the problem are displayed in Figure 15. The unique CO1 layout is exhibited in Figure 16. Figures 17 and 18 present the L-shaped and Orgun Pipe layouts for the problem. As before, both the layouts fail to better the CO1 layout, thereby validating the hypothesis. A small set of examples has been presented to illustrate the empirical insights gained into the dualaddress CO1 proof established in Theorem 1. These examples represent a small subset of the numerous examples that have been tested to verify the ideas presented here. An important observation resulting from the empirical study is that although the COI-based assignment yields multiple layouts, it still remains the optimal layout; the optimal layout in such a case can be identified by using the heuristic. Theoretical basis for the tie-breaking strategy

7.9126669409E+OO
Figure 16. COI layout and cost value for Problem 3

Consider the square rack face shown in Figure 19. Let an isorec be defined as a line connecting the locations that are at the same rectilinear distance from the I/O point. These isorectilinear distance lines are the diagonally oriented lines shown in Figure 19. A tie represents the possibility of locating an item at any one of the lattice points along an isorec. Recall that the tiebreaking strategy involves computation of the euclidean distances of the candidate locations from the I/O point and picking the one that is closest to the I/O point. In the event that a tie exists even after execution of the tie-breaking strategy, the following action is suggested on the basis of the shape of the rack face:
l

3
l

If the rack face is square, that is, has equal number of rows and columns in the grid, it does not matter which one of the tie locations is picked after the application of the tie-breaking strategy. If the rack face is not square, pick the location that is on the side of the principal axis of symmetry, that is, the leading diagonal shown in Figure 19, that contains the centroid of the rack face.

7.942270916E+00
Figure 17. L-shaped configuration for Problem 3

The square rack face case is considered initially. To understand the basis for the tie-breaking strategy, consider a single isorec with three lattice points on it as

Appl. Math. Modelling,

1990, Vol. 14, February

93

Revised proof of opiimality

of COI assignment

policy: C. J. Malmborg

and K. Bhaskaran

principal axis

isorec

virtue of the lemma). The perpendicular bisector of the isorec is the principal axis of symmetry or the leading diagonal. Hence the euclidean tie-breaking strategy locates the chosen lattice point on this axis of symmetry or closest to it. This has the effect of minimizing the component l,, in the dual-address orderpicking cost expression, where b and a are any two locations and lha is the rectilinear distance between these two locations. With respect to Figure 20, among the three points 1, 2, and 3, the most favorable point to minimize the lb0 values is 2, and it is the point chosen by the tie-breaking scheme. Consider the expected dualaddress order-picking cost, recalled from Theorem 1:

Figure 19.

Rack face illustrating

the isorecs b i

Llfh

I/O
Figure 20. Illustrative sketch for proof of the lemma

shown in Figure 20. The following lemma then provides insight into the theoretical reasoning.
Le#WM

The perpendicular dropped from the l/O point 0 to the isorec AB bisects the isorec, that is, if OC is perpendicular to AB, then AC = BC.
Proof OB = OA

The fraction (fjX,/Qj) is the same for all tie locations. The rectilinear distance from I/O point, ftTh, is the same for all the tie locations. The distance I<,(,, is the distance from all remaining locations to the I/O point, and this remains unaltered by the choice of any one of the tie locations. Hence the only cost component that is affected by the tie-breaking strategy is lbu. We have already seen how the euclidean tie-breaking strategy minimizes the component l,, among candidate layouts. This explains why the tie-breaking strategy picks the optimal CO1 layout from the set of CO1 layouts. So far the discussion has assumed a square rack face. For a rectangular rack face, which is not square in the number of rows and columns, the euclidean tiebreaking strategy works with a minor additional rule, as indicated at the outset. Consider the 3 x 5 rack face shown in Figure 21. The euclidean tie-breaking strategy would pick points A, B, C, and D when a tie occurred in any one of the corresponding isorecs. The reasoning is clear when the missing rack structure, shown by dashed lines in Figure 21, that would make it symmetric is considered. Points A and C lie on the principal axis of symmetry and are hence preferred. Point D is the closest location to the principal axis that

(by property of the square rack) (by construction) (sum of angles in a triangle is ISOO) (using triangle congruency)

angle OCB = OCA = 90


angle AOC = BOC

Hence n OAC = D OBC

:. AC = BC Computing the euclidean distance between 0 and the lattice points 1, 2, and 3 and picking the point closest to the I/O point 0 are equivalent to finding the shortest distance between point 0 and the isorec AB. The shortest distance from any point to a line is the perpendicular dropped from that point on to the line. Therefore computing the euclidean distances and breaking the tie using the minima ensure that the chosen lattice point lies on the perpendicular bisector of the isorec (by

Figure 21.

Tie-breaking

strategy for a rectangular

rack

94

Appl.

Math. Modelling,

1990, Vol. 14, February

Revised proof of optimality

of CO1 assignment
7 8

policy: C. J. Malmborg

and K. Bhaskaran

is in the rack and is therefore chosen. Points B and E lie equidistant with respect to the principal axis of symmetry. However, the tie-breaking strategy would pick B in preference to E, since it lies on the side of the principal axis that contains the centroid of the rack face. In fact, in this example, B is the centroid of the rack face. This addendum to the tie-breaking strategy is intuitive. When a portion of any square object is removed, the center of gravity shifts toward the heavier side of the object. The tie-breaking scheme for rectangular racks tries to reconcile with this tenet. Summary and conclusions

10

II 12 I3

Wilson, H. G. Order quantity, product popularity, and the location of stock in a warehouse. AIIE Trans. 9(3). 1977.230-237 Malmborg, C. J. and Bhaskaran. K. Optimal storage assignment for multiaddress warehousing systems. IEEE Trrozs. SMC 1989, 19(l), 197-205 Malmborg, C. J. and Bhaskaran, K. On the optimality of the cube per order index for conventional warehouses with dual command cycles. Muferial Now 1987, 4, 169-175 Bhaskaran, K. Modeling and analytical foundations of the stochastic, multiple objective optimization problem for integrated warehouse operations. PhD. Dissertation, Rensselaer Polytechnic Institute, Troy, New York, 1988 Bozer, Y. A. Personal correspondence. Dept. of IOE, Univ. of Michigan, Ann Arbor, Mich.. Feb. 1988 Hardy, G. H., Littlewood, J. E. and Polya. G. Inequulities. Cambridge University Press, London, 1952 Wong, C. K. A/,qorithmic Studies in Mass Storrrge System. Computer Science Press, 1983

A revised proof of optimality of the cube-per-order index assignment policy for dual-address systems has been presented. The empirical insights into the optimality of CO1 reveal the nonuniqueness of the layout. A heuristic based on a euclidean tie-breaking strategy has been provided to pick the optimal CO1 layout from the set of CO1 layouts. The study concludes that when the CO1 assignment policy yields more than one layout, it contains the optimal layout in the set of layouts, and in the event that it yields a unique layout, this is the optimal layout for the dual-address model described in Ref. 9. References
Hausman, W. H., Schwarz, L. B. and Graves. S. C. Optimal storage assignment in automatic warehousing systems. Management Sri. 1976, 22,629-638 Heskitt, J. L. Putting the CO1 to work in warehouse layout. Trunsportntion and Distribution Muncrgement 1964, 4, 23-30 Malmborg. C. J.. Balachandran, S. and Kyle, D. M. A model based evaluation of a commonly used rule of thumb for warehouse layout. Appl. Moth. Model/kg 1986. 10, 133-138 Harmatuck, D. J. A comparison of two approaches to stock location. The Logistks and Trunsportrrfiotr Rcvitw 1976, 12(4), 282-285 Kallina, C. and Lynn, J. Application of the rule for stock location in a distribution warehouse. Interfmce.\ 1976, 7(l), 37-45 Rosenblatt, M. J. and Roll, Y. Warehouse design for storage policy considerations. Internut. J. Production Rrs. 1984,22(5). 809-82 I

Appendix:

Organ Pipe Configuration

The Orgun Pipe configuration is based on the classical mathematical work of Hardy, Littlewood, and Polya.* Their concept is used to minimize the expected distance travelled by an access head as it travels from one musical record to another: If PI. p2. . . . p,, represents
place the record with the access probabilities, point. then

Then relatively place the next largest p,, alternating between the position immediately to the left (right) of those already
largest p, at any and the position immediately to the right (left).

placed

The typical assumptions underlying the above configuration in the context of mass storage systems include: I3

1. 2. 3. 4.

linear storage, records allocated at integral points, first in-first out queue requests, head stops at the position of the record cessed, and 5. independent access probabilities.

last ac-

It may be noted that some of these assumptions are at variance with those of the dual-address model discussed in Ref. 9.

Appl.

Math. Modelling,

1990, Vol. 14, February

95

Das könnte Ihnen auch gefallen