Project12 by Group05

Bin packing heuristics

Group05

Date: 2009.1.15

Project12 by Group05

Content
Chapter 1: Introduction 3
Chapter 2: Data Structure / Algorithm Specification....4
Next Fit ..4
First Fit 5
Best Fit.8
First Fit Decreasing.9
Chapter 3: Testing Results...10
Time Complexity10
Bin number...15
Chapter 4: Analysis and Comments16
Appendix: Source Code.........23
References..46
Athor list.....47
Declaration....47
Signatures..............................................47

Project12 by Group05

Chapter 1:

Introduction

This project requires us to implement and compare the performance

(both in time and number of bins used) of the various bin packing
heuristics, including the on-line, next fit, first fit, best fit, and first-fit
decreasing algorithms.
As we all know, in computational complexity theory, the bin packing
problem is a combinatorial NP-hard problem. In it, objects of different
volumes must be packed into a finite number of bins of capacity V in a
way that minimizes the number of bins used.
On the purpose of this report, we should compile different programs
to run those four algorithms.
Our results give a thorough show of the performance of the four
algorithms. The first fit algorithm provides a fast but often nonoptimal
solution, involving placing each item into the first bin in which it will fit.
It requires (N log N) time, where n is the number of elements to be
packed. The first decreasing algorithm can be made much more effective
by first sorting the list of elements into decreasing order. And for longer
lists may increase the running time of the algorithm. Next-fit algorithm
differs from first-fit algorithm in that a first-fit algorithm places each
item into the first bin in which it will fit, whereas a next-fit algorithm just
check whether a new item fits in the same bin as the last item and just
3

Project12 by Group05

create a new bin if not, and is significantly faster than the first-fit. Of
course, it uses more bins. Best fit is the algorithm places objects in the
tightest spot among all bins. It perform better for random inputs and
needs O( NlogN) time.

Chapter 2:

Data Structure / Algorithm Specification

In this project, we use different data structures to implement the

four bin packing algorithms, aiming to get lower time complexity. Here,
we will introduce the data structures used in the four algorithms as well
as how the algorithms work.
The meaning of the inputs in the pseudo codes is shown as follows:
Num: the number of items to packing in the bins;
Item[]: an array to store the weight of each item;
C: the contain of the bins.

Algorithm 1:

Next Fit (on-line)

Definition: When processing any item, we check to see whether it

fits in the same bin as the last item. If it does, it is placed there;
otherwise, a new bin is created.

Project12 by Group05

Data Structure: To implement the algorithm in linear time

complexity is easy. Therefore, we just use some arrays and loops to
implement it.
Pseudo Code of Next Fit:
int NextFit (int Num, int Item[], int C)
{
int bin[Num]; //an array to record the contain of each bin, Num items
need at most Num bins
Initial each bins contain to zero at the beginning;
for(each bin[i]){
for(each Item[j]){
if (Item[j] can be fit in bin[i])
bin[i]=bin[i]+Item[j]; //pack Item[j] in bin[i] and process next
item
else (Item[j] cannot be fit in bin[j])
break; //go to the next bin to see if it can fit Item[j]
}
}
bin_num = the number of bins whose contains are not zero;
return bin_num;
}

Algorithm 2: First Fit (on-line)

Definition: To scan the bins in order and place the new item in the
first bin that is large enough to hold it. Thus, a new bin is created only

Project12 by Group05

when the results of the previous placements have left no other

alternative.
Data structure: to implement the First Fit algorithm in O(NlogN) time
complexity, we use a new data structure Max Winner Tree. The
introduction of the Winner Tree is shown as follows:
Winner Tree:
Declaration:
- A winner tree for n players is a complete binary tree with n
external nodes and n-1 internal nodes. Each internal node
records the winner of each match. To determine the winner, we
assume that each player has a value. In a max(min) winner tree,
the player who has th larger(smaller) value wins.
- An array t[1] to t[n-1] to represent the internal nodes;
- An array e[1] to e[n] to represent the players.
Here are two examples of the winner tree:

Min winner tree

Max winner tree

6

Project12 by Group05

Operation:
- Creation(n): to create a winner tree with at most n players;
- Initialize(player[]): to insert the players in the external nodes
and play the the whole game.
- Replay: when the value of one player has changed, the game
has to been replayed.
- Winner(a,b): to decide the winner of player a and player b;
Advantages:
When the value of one player has changed, it is easy and fast to
replay the game in O(logN) time. More detailed analysis of the time
complexity is shown in Chapter 4.
Pseudo Code of First Fit:
Int FirsttFit (int Num, int Item[], int C)
{
CreateAWinnerTree(Num);
Initialize(bin[]=C); //the bins are the players and the available space is the value
of each player
for(Item[i]){
start from the leftchild of the root T[Q];
do {
if (Item[i] can be fit in bin[T[Q]])
go to the leftchild of T[Q] and go on checking;
else (Item[i] cannot be fit in bin[T[Q]]){
7

Project12 by Group05

go to the rightchild of T[Q] and go on checking;

} until(T[Q] has no child);
//after the while loop above, we will finally find the first bin that can fit
Item[i], then we pack the Item[j] in the bin
//bin[T[Q]] is the first fit bin found
bin[T[Q]]=bin[T[Q]] Item[i]; //packing the item in the bin
Replay the game and go on packing the next Item;
}
bin_num = the number of bins whose contains are not zero;
return bin_num;
}

Algorithm 3: Best Fit (on-line)

Definition: To scan the bins in order and place the new item in the
tightest bin that is large enough to hold it. Thus, a new bin is created
only when the results of the previous placements have left no other
alternative.
Data Structure: to implement the algorithm in O(NlogN) time
complexity, we use a data structure we have learned before. That is
Splay Tree. Since we have learned this data structure before, here we
dont introduce the Splay Tree explicitly.
Pseudo Code of Best Fit:
int BeststFit(int Num, int Item[],int C)
{
int bin_num = 0; //record the bin used
8

Project12 by Group05

SplayTree

Tree; //the bins are stored in a splay tree

for(each Item[i]){
if(Tree==NULL){//no bins in the tree
Create a bin in the Tree and pack Item[i] in it;
bin_num++;
continue; //go on packing the next bin
}
find the bin in the tree whose space is closest to Item[i];
if(Item[i] cannot fit in the bin found){
Create a bin in the Tree and pack Item[i] in it;
bin_num++;
continue; //go on packing the next Item[i]
}
else if (Item[i] exactly fits the bin found)
Delete the bin from the Tree ; //since the bin is full
else if (the bin found is a little larger the Item[i]){
pack the Item[i] in the bin found and Update the Tree;
continue; //go to the next item;
}
}
return bin_num;
}

Algorithm 4: First Fit Decreasing (off-line)

Definition: we scan all the weight of the item and sort them in
decreasing order, and then we employ the First Fit algorithm with the
sorted items.
9

Project12 by Group05

10

Data Structure: to implement this algorithm, we simply sort the

items with QuickSort algorithm, and then use the First Fit algorithm.
Pseudo Code of First Fit Decreasing:
int FirstFitDecreasing(int Num, int Item[],int C)
{
QuickSort(Item[]);//sort the items in decreasing order
bin_num=FirstFit(Num, Item[], C);//employ First Fit algorithm
return bin_num;
}

Chapter 3:

Testing Results

In this chapter, we use several test cases to test the running time and
bin number of the four different algorithms. Whats more, we also test
the bin contain level of each algorithm, the bin contain level refers to
how full the bin is. Here, the specific testing input data is not included in
this report but will be contained in a file called Test.

Testing results for time complexity

Total Number: the total number of items to be packed;
Input: the weight of each item;
Case 1:
When the input is in random order, we test the running time of the
four algorithms in different total number.
The testing result is shown in Figure 1 as follows:
10

Project12 by Group05

Figure 1

11

( unit: second)

TOTAL NUMBER
100
500
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
11000
12000
13000
14000

NEXTFIT

FIRSTFIT

0.000005
0.000019
0.000034
0.000066
0.000103
0.000131
0.000164
0.000206
0.000240
0.000274
0.000295
0.000328
0.000378
0.000394
0.000446
0.000459

0.000045
0.000273
0.000594
0.001314
0.002066
0.002812
0.003670
0.004422
0.005176
0.006000
0.006901
0.007970
0.008678
0.009602
0.010289
0.011225

BESTFIT

FIRSTFITDECREASING

0.000048
0.000266
0.000547
0.001062
0.001643
0.002188
0.002735
0.003295
0.003852
0.004376
0.004928
0.005470
0.005900
0.006590
0.007184
0.007704

0.000052
0.000304
0.000656
0.001438
0.002297
0.003060
0.004140
0.004898
0.005718
0.006504
0.007604
0.008590
0.009900
0.010542
0.011513
0.012324

Case 2:
When the input is in decreasing order, we test the running time of
the four algorithms in different total number.
The testing result is shown in Figure 2 as follows:
Figure 2

TOTAL NUMBER
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000

( unit: second)

NEXTFIT
0.000028
0.000053
0.000075
0.000106
0.000125
0.000150
0.000197
0.000212
0.000239
0.000266

FIRSTFIT
0.000594
0.001282
0.002066
0.002752
0.003675
0.004428
0.005282
0.005880
0.006892
0.00797

BESTFIT
0.002765
0.009218
0.022192
0.038748
0.065855
0.094880
0.125662
0.176000
0.226919
0.29109

FIRSTFITDECREASING
0.000656
0.001406
0.002300
0.003000
0.004065
0.004892
0.005831
0.006496
0.007604
0.00859

11

Project12 by Group05

12

Case 3:
When the input is in increasing order, we test the running time of the
four algorithms in different total number.
The testing result is shown in Figure 3 as follows:
Figure 3

( unit: second)

TOTALNUMBER
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000

NEXTFIT

FIRSTFIT

BESTFIT

0.000030
0.000053
0.000080
0.000100
0.000133
0.000150
0.000186
0.000200
0.000239
0.000265

0.000563
0.001220
0.001970
0.002624
0.003515
0.004235
0.005063
0.005752
0.006622
0.007500

0.000656
0.001624
0.003003
0.005252
0.007345
0.010169
0.013535
0.016000
0.021676
0.025780

FIRSTFITDECREASING
0.000641
0.001406
0.002300
0.003000
0.004140
0.004988
0.005718
0.006496
0.007459
0.008600

Case 4:
When the input is in different order, we test the running time of the
Next Fit algorithm in different total number.
The testing result is shown in Figure 4 as follows:
Case 5:
When the input is in different order, we test the running time of the First
Fit algorithm in different total number.
The testing result is shown in Figure 5 as follows:
12

Project12 by Group05

Figure 4

TOTAL NUMBER
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000

(unit: second)

NEXTFIT
INCREASING
0.000030
0.000053
0.000080
0.000100
0.000133
0.000150
0.000186
0.000200
0.000239
0.000265

NEXTFIT
DECREASING
0.000028
0.000053
0.000075
0.000106
0.000125
0.000150
0.000197
0.000212
0.000239
0.000266

Figure 5

TOTAL NUMBER
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000

13

NEXT FIT
RANDOM
0.000034
0.000066
0.000103
0.000131
0.000164
0.000206
0.000240
0.000274
0.000295
0.000328

( unit: second)

FIRSTFIT
INCREASING
0.000563
0.001220
0.001970
0.002624
0.003515
0.004235
0.005063
0.005752
0.006622
0.007500

FIRSTFIT
INCREASING
0.000594
0.001282
0.002066
0.002752
0.003675
0.004428
0.005282
0.005880
0.006892
0.00797

FIRSTFIT
RANDOM
0.000594
0.001314
0.002066
0.002812
0.003670
0.004422
0.005176
0.006000
0.006901
0.007970

Case 6:
When the input is in different order, we test the running time of the Best
Fit algorithm in different total number.
The testing result is shown in Figure 6 as follows:
13

Project12 by Group05

Figure 6

14

( unit: second)

TOTAL NUMBER
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000

BESTFIT
INCREASING
0.000656
0.001624
0.003003
0.005252
0.007345
0.010169
0.013535
0.016000
0.021676
0.025780

BESTFIT
DECREASING
0.002765
0.009218
0.022192
0.038748
0.065855
0.094880
0.125662
0.176000
0.226919
0.29109

BESTFIT
RANDOM
0.000547
0.001062
0.001643
0.002188
0.002735
0.003295
0.003852
0.004376
0.004928
0.005470

Case 7:
According to our previous analysis, we thought that the size of the
bin may affect the running time of the Best Fit algorithm. Here, we
provide a test for Best Fit, to compare the running time of Best Fit when
bin size is 10e7 and 10, respectively.
The testing result is shown in Figure 7 as follows:
Figure 7

(unit: second)

TOTALNUMBER
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000

BESTFIT
1E7
0.000547
0.001062
0.001643
0.002188
0.002735
0.003295
0.003852
0.004376
0.004928
0.00547

BESTFIT
10
0.000203
0.000374
0.000517
0.000752
0.00094
0.001133
0.001324
0.001496
0.001685
0.00188
14

Project12 by Group05

15

Testing results for Bin Number

Case 8:
When total number = 100000, we test the number of bins of the four
algorithms with input in random order and increasing order respectively.
This test case is to compare the performance of the four algorithms in
the same level. Also compare them with the ideal number of bins
needed.
The testing result is shown is Figure 8 as follows:
Figure 8

ORDER
Random
Increasing

NEXT FIT

FIRSTFIT

66890
64598

50970
64598

BESTFIT
50709
64598

FIRSTFITDEC
REASING
50426
50519

IDEAL
50401
50444

Case 9:
In different total numbers, we test the bin number of the four
algorithms and also compare with the idea number of bin needed.
Furthermore, we also test the bin contain level of each algorithm, that is,
how full is each bin on average.
The testing result is shown is Figure 9 as follows:
In the following Figure, the data with % refers to the bin contain level,
and data beside is the bin number.
15

Project12 by Group05

16

Figure 9

TOTAL NUMBER
10
100
1000
10000
100000

NEXTFIT
88.80%
5
75.65%
62
75.68% 684
75.31% 6711
75.31% 66815

Chapter 4:

FIRSTFIT
88.80%
5
91.96%
51
97.46% 555
98.89% 5211
98.89% 50880

BESTFIT
88.80%
5
91.96%
51
98.29% 547
99.29% 5167
99.29% 50673

FIRSTFITDECREASING IDEAL
88.80%
5
5
97.71%
48
47
99.62%
543
523
99.78%
5098
5079
99.78%
50428
50316

Analysis and Comments

Analysis of Case 1, Case 2 and Case 3:

According to the testing result in Figure 1, Figure 2 and Figure3, we
get the following Diagram 1, Diagram 2, Diagram 3 and Diagram4, which
are showing the running time of the four algorithms with random input,
decreasing input and increasing input.
Diagram 1 : Random Input

0.014000
0.012000
0.010000

NextFit

0.008000

FirstFit

0.006000

BestFit

0.004000

FirstFitDecreasing

0.002000

0.000000

16

Project12 by Group05

17

Diagram 2: Decreasing Input

0.350000
0.300000
0.250000

NextFit

0.200000

FirstFit

0.150000

BestFit

0.100000

FirstFitDecreasing

0.050000

0.000000

Diagram 3 : Decreasing Input

0.010000
0.009000
0.008000
0.007000
0.006000
0.005000
0.004000
0.003000
0.002000
0.001000
0.000000

NextFit
FirstFit
FirstFitDecreasing

Diagram 4 : Increasing Input

0.030000
0.025000

0.020000
0.015000
0.010000

NextFit
FirstFit
BestFit

FirstFitDecreasing

0.005000

0.000000

17

Project12 by Group05

18

The four diagrams above give us a more intuitively view of the

performance of the four algorithms. We can see that , no matter in
which input order, the Next Fit always runs the fastest, and much faster
than the other three; the First Fit runs nearly the same with the First Fit
Decreasing.
Analysis of Case 4, Case 5 and Case 6:
According to the testing result in Figure 4, Figure 5 and Figure 6, we
get the following Diagram 5, Diagram 6, Diagram 7, which are showing
the running time of Next Fit, First Fit and Best Fit with input in different
order respectively.
Diagram 5: Next Fit

0.000350
0.000300
0.000250

0.000200
0.000150

NextFit increasing
NextFit decreasing
Next fit random

0.000100
0.000050
0.000000

From Diagram 5 above, we see that Next Fit runs almost the same
when the input is in increasing order and decreasing order, but runs
much more slowly when the input is in random order.

18

Project12 by Group05

19

Diagram 6 : First Fit

0.009000
0.008000
0.007000
0.006000

FirstFit increasing

0.005000

FirstFit decreasing

0.004000

firstfit random

0.003000
0.002000
0.001000
0.000000

From the Diagram 6 above, we see that First Fit runs almost the same
no matter what order in the input.
Diagram 7 : Best Fit

0.350000
0.300000

BestFit increasing

0.250000
0.200000

BestFit decreasing

0.150000
BestFit random

0.100000
0.050000
10000

9000

8000

7000

6000

5000

4000

3000

2000

1000

0.000000

Form Diagram 7 above, we see that Best Fit runs the fastest when the
input is in random and runs a little slowly when the input is in increasing
order. And when the input is in decreasing order, Best Fit runs much
really slowly.
19

Project12 by Group05

20

Analysis of Case 7:
According to the testing result in Figure 7, we get the following
Diagram 8, which is showing the running time of Best Fit when the bin
size is 10e7 and 10 respectively.
Diagram 8 : Best Fit with different bin size

0.006
0.005
0.004
0.003

BestFit 1e7
BestFit 10

0.002

0.001
0

As what we have argued before, how fast the Best Fit runs is related
to the bin size. From the Diagram 8 above, we see that , the larger the
bin size is ,the more slowly Best Fit runs.

Analysis of Case 8 and Case 9:

According to the testing result in Figure 8 and Figure 9, we get the
following Diagram 9, which is showing the bin contain level of the four
algorithms with different total number of items.

20

Project12 by Group05

21

Diagram 9 : Bin contain level

120.00%
100.00%

80.00%
NextFit
60.00%

FirstFit
BestFit

40.00%

FirstFitDecreasing

20.00%

0.00%
10

100

1000

10000

100000

From the Diagram 9, we see that when the total number is small, the
average bin contain level of the four algorithms is almost the same. But
when the total number get larger, the difference is more apparent, with
that First Fit Decreasing is the best and Next Fit is the worst. The results
also reflect that, with the same total weight of all the items, the First Fit
Decreasing use the fewest number of bins and the second fewest is Best
Fit, and then the First Fit. The Next Fit use the most bins.
Conclusion:
Next Fit
In the next fit program, we use two for loops. And when all the
objects are put in the bins and z, which we use to count the number of
those objects in the bins, increases to n. In this case, we can never come
into the second for loop again, so we can see the results of our test
21

Project12 by Group05

22

program, the running time is O(N), N represents the number of the

objects. As seen in the figure, the next fit program use more bins, in
other words, the bins are not so full compared with other programs.
That is because we just create a new bin if the objects are too bigger.
We never check there are enough space in the front bins except the last
one.
First Fit
We use data type winner tree to build our first fit algorithm. As seen
in the source code, winner tree is a full binary tree, we just check from
the root to find bins could load our objects.
We ensure it is the first available bin to be used in our code. we just
need to change the information on the node on the way from the leaf
node which changes values to the root .In this way ,we get a O(NlogN)
time complexity. our test data prove it .Of course ,it uses more bins than
best fit, and less than next fit.
Best Fit
Best fit implemented with Splay tree runs in O(

) if the sizes of

the elements are all different from each other. When the total number
of sizes kinds are far larger than the number of the elements, time
complexity of this program is O(

).Otherwise, if the total number of

size kinds are far smaller than the number of the elements, time
complexity of this program is near O(N).
22

Project12 by Group05

23

So when the total number of size kinds is small, Best fit performs
much better than First fit on time.
As it is seen in the figure, best fit use less bins than first fit.
First Fit Decreasing
First fit decreasing is similar to first fit algorithm. But the difference is ,
the objects must be sorted firstly. So, it uses more time, but because the
quick sort does not affect the complexity, it still just needs O(NlogN)
time. And because the objects are sorted firstly, it needs the fewest bins.

Appendix:

Source Code (in C )

1. WinnerTree.c
#include<stdio.h>
#include<stdlib.h>

struct WinnerTree;
typedef struct WinnerTree* WTree;

WTree CreatWTree(int TreeSize);

void Initialize(WTree WT, int a[],int Size);
void Play(WTree WT,int k, int lc, int rc);
void RePlay(WTree WT, int k);
int

Winner(int a[], int b, int c);

int

WinnerT(WTree WT, int m);

struct WinnerTree
23

Project12 by Group05

46

fdr=FirstFitDecreasing(Total,S,BinSize);
stop = clock();
tFD=((double)(stop - start))/CLK_TCK;
tFD/=FFD_repeat;

/*calculate the average contain of each bin */

dstN=sum/(double)nr;
dstF=sum/(double)fr;
dstB=sum/(double)br;
dstFD=sum/(double)fdr;

/*print the testing results*/

printf("NextFit: %d
printf("FirstFit: %d
printf("BestFit: %d

AVG:%lf
AVG:%lf
AVG:%lf

printf("FirstFitDecreasing: %d

Total time:%lf\n",nr,dstN,tN);
Total time:%lf\n",fr,dstF,tF);
Total time:%lf\n",br,dstB,tB);
AVG:%lf Total time:%lf\n",fdr,dstFD,tFD);

return 1;
}

References:
[1]Mark Allen Weiss, Data Structures and Algorithm Analysis in C,
POSTS&TELECOM PRESS, 2005.
[2] http://en.wikipedia.org/wiki/Winner_tree
[3] http://en.wikipedia.org/wiki/Splay_tree

46

Project12 by Group05

47

Author Lists:
Programmer:
Report Writer:
Tester
PPT maker

Declaration:
We hereby declare that all the work done in this project titled
"Bin Packing Heuristics" is of our independent effort as a
group.

Signatures:

47