First-fit-decreasing bin packing

<h2 id="description">Description</h2>
The FFD algorithm works as follows.

<ul><li>Order the items from largest to smallest.</li>
<li>Open a new empty bin, bin #1.</li>
<li>For each item from largest to smallest, find the first bin into which the item fits, if any.
<ul><li>If such a bin is found, put the new item in it.</li>
<li>Otherwise, open a new empty bin put the new item in it.</li></ul></li></ul>
In short: FFD orders the items by descending size, and then calls <a href="/facts/First-fit_bin_packing/1YpJDIg8">first-fit bin packing</a>. 
An equivalent description of the FFD algorithm is as follows. 

<ul><li>Order the items from largest to smallest.</li>
<li>While there are remaining items:
<ul><li>Open a new empty bin.</li>
<li>For each item from largest to smallest:
<ul><li>If it can fit into the current bin, insert it.</li></ul></li></ul></li></ul>
In the standard description, we loop over the items once, but keep many open bins. In the equivalent description, we loop over the items many times, but keep only a single open bin each time.

<h2 id="performance-analysis">Performance analysis</h2>
The performance of FFD was analyzed in several steps. Below, 
 
 
 
 F
 F
 D
 (
 S
 ,
 C
 )
 
 
 {\displaystyle FFD(S,C)}
 
 denotes the number of bins used by FFD for input set S and bin-capacity C.

<ul><li>In 1973, <a href="/facts/David_S._Johnson/kGSBxucb">D.S. Johnson</a> proved in his doctoral thesis<a class="footnote-ref" id="fnref:1" href="#fn:1">1</a> that 
 
 
 
 F
 F
 D
 (
 S
 ,
 C
 )
 ≤
 11
 
 /
 
 9
 
 O
 P
 T
 
 (
 S
 ,
 C
 )
 +
 4
 
 
 {\displaystyle FFD(S,C)\leq 11/9\mathrm {OPT} (S,C)+4}
 
 for any instance S and capacity C.</li>
<li>In 1985, B.S. Backer<a class="footnote-ref" id="fnref:2" href="#fn:2">2</a> gave a slightly simpler proof and showed that the additive constant is not more than 3.</li>
<li>Yue Minyi<a class="footnote-ref" id="fnref:3" href="#fn:3">3</a> proved that 
 
 
 
 F
 F
 D
 (
 S
 ,
 C
 )
 ≤
 11
 
 /
 
 9
 
 O
 P
 T
 
 (
 S
 ,
 C
 )
 +
 1
 
 
 {\displaystyle FFD(S,C)\leq 11/9\mathrm {OPT} (S,C)+1}
 
 in 1991 and, in 1997, improved this analysis to 
 
 
 
 F
 F
 D
 (
 S
 ,
 C
 )
 ≤
 11
 
 /
 
 9
 
 O
 P
 T
 
 (
 S
 ,
 C
 )
 +
 7
 
 /
 
 9
 
 
 {\displaystyle FFD(S,C)\leq 11/9\mathrm {OPT} (S,C)+7/9}
 
 together with Li Rongheng.<a class="footnote-ref" id="fnref:4" href="#fn:4">4</a></li>
<li>In 2007 György Dósa<a class="footnote-ref" id="fnref:5" href="#fn:5">5</a> proved the tight bound 
 
 
 
 F
 F
 D
 (
 S
 ,
 C
 )
 ≤
 11
 
 /
 
 9
 
 O
 P
 T
 
 (
 S
 ,
 C
 )
 +
 6
 
 /
 
 9
 
 
 {\displaystyle FFD(S,C)\leq 11/9\mathrm {OPT} (S,C)+6/9}
 
 and presented an example for which 
 
 
 
 F
 F
 D
 (
 S
 ,
 C
 )
 =
 11
 
 /
 
 9
 
 O
 P
 T
 
 (
 S
 ,
 C
 )
 +
 6
 
 /
 
 9
 
 
 {\displaystyle FFD(S,C)=11/9\mathrm {OPT} (S,C)+6/9}
 
.</li></ul>
<h3>Worst-case example</h3>
The lower bound example given in by Dósa is the following: Consider the two bin configurations:

<ul><li>
 
 
 
 
 B
 
 1
 
 
 :=
 {
 1
 
 /
 
 2
 +
 ε
 ,
 1
 
 /
 
 4
 +
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 }
 
 
 {\displaystyle B_{1}:=\{1/2+\varepsilon ,1/4+\varepsilon ,1/4-2\varepsilon \}}
 
  ;</li>
<li>
 
 
 
 
 B
 
 2
 
 
 :=
 {
 1
 
 /
 
 4
 +
 2
 ε
 ,
 1
 
 /
 
 4
 +
 2
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 }
 
 
 {\displaystyle B_{2}:=\{1/4+2\varepsilon ,1/4+2\varepsilon ,1/4-2\varepsilon ,1/4-2\varepsilon \}}
 
 .</li></ul>
If there are 4 copies of 
 
 
 
 
 B
 
 1
 
 
 
 
 {\displaystyle B_{1}}
 
 and 2 copies of 
 
 
 
 
 B
 
 2
 
 
 
 
 {\displaystyle B_{2}}
 
 in the optimal solution, FFD will compute the following bins: 

<ul><li>4 bins with configuration 
 
 
 
 {
 1
 
 /
 
 2
 +
 ε
 ,
 1
 
 /
 
 4
 +
 2
 ε
 }
 
 
 {\displaystyle \{1/2+\varepsilon ,1/4+2\varepsilon \}}
 
,</li>
<li>1 bin with configuration 
 
 
 
 {
 1
 
 /
 
 4
 +
 ε
 ,
 1
 
 /
 
 4
 +
 ε
 ,
 1
 
 /
 
 4
 +
 ε
 }
 
 
 {\displaystyle \{1/4+\varepsilon ,1/4+\varepsilon ,1/4+\varepsilon \}}
 
,</li>
<li>1 bin with configuration 
 
 
 
 {
 1
 
 /
 
 4
 +
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 }
 
 
 {\displaystyle \{1/4+\varepsilon ,1/4-2\varepsilon ,1/4-2\varepsilon ,1/4-2\varepsilon \}}
 
,</li>
<li>1 bin with configuration 
 
 
 
 {
 1
 
 /
 
 4
 −
 2
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 ,
 1
 
 /
 
 4
 −
 2
 ε
 }
 
 
 {\displaystyle \{1/4-2\varepsilon ,1/4-2\varepsilon ,1/4-2\varepsilon ,1/4-2\varepsilon \}}
 
,</li>
<li>1 one final bin with configuration 
 
 
 
 {
 1
 
 /
 
 4
 −
 2
 ε
 }
 
 
 {\displaystyle \{1/4-2\varepsilon \}}
 
,</li></ul>
That is, 8 bins total, while the optimum has only 6 bins. Therefore, the upper bound is tight, because 
 
 
 
 11
 
 /
 
 9
 ⋅
 6
 +
 6
 
 /
 
 9
 =
 72
 
 /
 
 9
 =
 8
 
 
 {\displaystyle 11/9\cdot 6+6/9=72/9=8}
 
. 
This example can be extended to all sizes of 
 
 
 
 
 OPT
 
 (
 S
 ,
 C
 )
 
 
 {\displaystyle {\text{OPT}}(S,C)}
 
:<a class="footnote-ref" id="fnref:6" href="#fn:6">6</a> in the optimal configuration there are 9k+6 bins: 6k+4 of type B1 and 3k+2 of type B2. But FFD needs at least 11k+8 bins, which is 
 
 
 
 
 
 11
 9
 
 
 (
 6
 k
 +
 4
 +
 3
 k
 +
 2
 )
 +
 
 
 6
 9
 
 
 
 
 {\displaystyle {\frac {11}{9}}(6k+4+3k+2)+{\frac {6}{9}}}
 
.

<h3>Performance with divisible item sizes</h3>
An important special case of bin-packing is that which the item sizes form a divisible sequence (also called factored). A special case of divisible item sizes occurs in memory allocation in computer systems, where the item sizes are all powers of 2. In this case, FFD always finds the optimal packing.<a class="footnote-ref" id="fnref:7" href="#fn:7">7</a>: Thm.2 

<h2 id="monotonicity-properties">Monotonicity properties</h2>
Contrary to intuition, 
 
 
 
 F
 F
 D
 (
 S
 ,
 C
 )
 
 
 {\displaystyle FFD(S,C)}
 
 is not a monotonic function of C. <a class="footnote-ref" id="fnref:8" href="#fn:8">8</a>: Fig.4  Similarly, 
 
 
 
 F
 F
 D
 (
 S
 ,
 C
 )
 
 
 {\displaystyle FFD(S,C)}
 
 is not a monotonic function of the sizes of items in S: it is possible that an item shrinks in size, but the number of bins increases.
However, the FFD algorithm has an "asymptotic monotonicity" property, defined as follows.<a class="footnote-ref" id="fnref:9" href="#fn:9">9</a>: Lem.2.1 

<ul><li>For every instance S and integer m, let MinCap(S,m) be the smallest capacity C such that 
 
 
 
 O
 P
 T
 (
 S
 ,
 C
 )
 ≤
 m
 
 
 {\displaystyle OPT(S,C)\leq m}
 
</li>
<li>For every integer m, let MinRatio(m) be the <a href="/facts/Infimum_and_supremum/oXCKVopz">infimum</a> of the numbers r≥1 such that, for all input sets S, 
 
 
 
 F
 F
 D
 (
 S
 ,
 r
 ⋅
 
 MinCap
 
 (
 S
 ,
 m
 )
 )
 ≤
 m
 
 
 {\displaystyle FFD(S,r\cdot {\text{MinCap}}(S,m))\leq m}
 
. This is the amount by which we need to "inflate" the bins such that FFD attains the optimal number of bins.</li>
<li>Then, for every input S and for every r ≥ MinRatio(m), 
 
 
 
 F
 F
 D
 (
 S
 ,
 r
 ⋅
 
 MinCap
 
 (
 S
 ,
 m
 )
 )
 ≤
 m
 
 
 {\displaystyle FFD(S,r\cdot {\text{MinCap}}(S,m))\leq m}
 
. This shows, in particular, that the infimum in the above definition can be replaced by minimum.</li></ul>
<h3>Examples</h3>
For example, suppose the input is: <blockquote>44, 24, 24, 22, 21, 17, 8, 8, 6, 6. </blockquote>With capacity 60, FFD packs 3 bins:
<ul><li>44, 8, 8;</li>
<li>24, 24, 6, 6;</li>
<li>22, 21, 17.</li></ul>
But with capacity 61, FFD packs 4 bins:

<ul><li>44, 17;</li>
<li>24, 24, 8;</li>
<li>22, 21, 8, 6;</li>
<li>6.</li></ul>
This is because, with capacity 61, the 17 fits into the first bin, and thus blocks the way to the following 8, 8.
As another example,<a class="footnote-ref" id="fnref:10" href="#fn:10">10</a>:  Ex.5.1   suppose the inputs are: 51, 28, 28, 28, 27, 25, 12, 12, 10, 10, 10, 10, 10, 10, 10, 10. With capacity 75, FFD packs 4 bins:

<ul><li>51, 12, 12</li>
<li>28, 28, 10</li>
<li>28, 27, 10, 10</li>
<li>25, 10, 10, 10, 10, 10</li></ul>
But with capacity 76, it needs 5 bins:

<ul><li>51, 25</li>
<li>28, 28, 12</li>
<li>28, 27, 12</li>
<li>10, 10, 10, 10, 10, 10, 10</li>
<li>10</li></ul>
Consider the above example with capacity 60. If the 17 becomes 16, then the resulting packing is:

<ul><li>44, 16;</li>
<li>24, 24, 8;</li>
<li>22, 21, 8, 6;</li>
<li>6.</li></ul>
<h2 id="modified-first-fit-decreasing">Modified first-fit-decreasing</h2>
Modified first fit decreasing (MFFD)<a class="footnote-ref" id="fnref:11" href="#fn:11">11</a> improves on FFD for items larger than half a bin by classifying items by size into four size classes large, medium, small, and tiny, corresponding to items with size > 1/2 bin, > 1/3 bin, > 1/6 bin, and smaller items respectively. Then it proceeds through five phases:

<ol><li>Allot a bin for each large item, ordered largest to smallest.</li>
<li>Proceed forward through the bins. On each: If the smallest remaining medium item does not fit, skip this bin. Otherwise, place the largest remaining medium item that fits.</li>
<li>Proceed backward through those bins that do not contain a medium item. On each: If the two smallest remaining small items do not fit, skip this bin. Otherwise, place the smallest remaining small item and the largest remaining small item that fits.</li>
<li>Proceed forward through all bins. If the smallest remaining item of any size class does not fit, skip this bin. Otherwise, place the largest item that fits and stay on this bin.</li>
<li>Use FFD to pack the remaining items into new bins.</li></ol>
This algorithm was first studied by Johnson and Garey<a class="footnote-ref" id="fnref:12" href="#fn:12">12</a> in 1985, where they proved that 
 
 
 
 M
 F
 F
 D
 (
 S
 ,
 C
 )
 ≤
 (
 71
 
 /
 
 60
 )
 
 O
 P
 T
 
 (
 S
 ,
 C
 )
 +
 (
 31
 
 /
 
 6
 )
 
 
 {\displaystyle MFFD(S,C)\leq (71/60)\mathrm {OPT} (S,C)+(31/6)}
 
. This bound was improved in the year 1995 by Yue and Zhang<a class="footnote-ref" id="fnref:13" href="#fn:13">13</a> who proved that 
 
 
 
 M
 F
 F
 D
 (
 S
 ,
 C
 )
 ≤
 (
 71
 
 /
 
 60
 )
 
 O
 P
 T
 
 (
 S
 ,
 C
 )
 +
 1
 
 
 {\displaystyle MFFD(S,C)\leq (71/60)\mathrm {OPT} (S,C)+1}
 
.

<h2 id="other-variants">Other variants</h2>
Best-fit-decreasing (BFD) is very similar to FFD, except that after the list is sorted, it is processed by <a href="/facts/Best-fit_bin_packing/JI3prfPs">best-fit bin packing</a>. Its asymptotic approximation ratio is the same as FFD - 11/9.

<h2 id="implementations">Implementations</h2>
<ul><li>Python: The <a href="https://github.com/erelsgl/prtpy">prtpy package</a> contains <a href="https://github.com/erelsgl/prtpy/blob/main/prtpy/packing/first_fit.py">an implementation of first-fit decreasing</a>.</li></ul>
<h2 id="see-also">See also</h2>
<ul><li><a href="/facts/Multifit_algorithm/41TkvqaW">Multifit algorithm</a> - an algorithm for <a href="/facts/Identical-machines_scheduling/lwNSXMs6">identical-machines scheduling</a>, which uses FFD as a subroutine.</li></ul>

<h2 id="references">References</h2>

<ol>
<li id="fn:1">Johnson, David S (1973). "Near-optimal bin packing algorithms" (PDF). Massachusetts Institute of Technology. <a href="https://dspace.mit.edu/bitstream/handle/1721.1/57819/17595570-MIT.pdf?sequence=2" target="_blank">https://dspace.mit.edu/bitstream/handle/1721.1/57819/17595570-MIT.pdf?sequence=2</a> <a href="#fnref:1" class="footnote-back-ref">↩</a></li>
<li id="fn:2">Baker, Brenda S. (1985). "A New Proof for the First-Fit Decreasing Bin-Packing Algorithm". J. Algorithms. 6 (1): 49–70. doi:10.1016/0196-6774(85)90018-5. <a href="/wiki/Doi_(identifier)" target="_blank">/wiki/Doi_(identifier)</a> <a href="#fnref:2" class="footnote-back-ref">↩</a></li>
<li id="fn:3">Yue, Minyi (October 1991). "A simple proof of the inequality FFD (L) ≤ 11/9 OPT (L) + 1, ∀L for the FFD bin-packing algorithm". Acta Mathematicae Applicatae Sinica. 7 (4): 321–331. doi:10.1007/BF02009683. S2CID 189915733. <a href="/wiki/Doi_(identifier)" target="_blank">/wiki/Doi_(identifier)</a> <a href="#fnref:3" class="footnote-back-ref">↩</a></li>
<li id="fn:4">Li, Rongheng; Yue, Minyi (August 1997). "The proof of FFD(L) < -OPT(L) + 7/9". Chinese Science Bulletin. 42 (15): 1262–1265. Bibcode:1997ChSBu..42.1262L. doi:10.1007/BF02882754. S2CID 93280100. <a href="/wiki/Bibcode_(identifier)" target="_blank">/wiki/Bibcode_(identifier)</a> <a href="#fnref:4" class="footnote-back-ref">↩</a></li>
<li id="fn:5">Dósa, György (2007). "The Tight Bound of First Fit Decreasing Bin-Packing Algorithm is FFD(I) ≤ 11/9OPT(I) + 6/9". In Chen Bo; Mike Paterson; Zhang Guochuan (eds.). Combinatorics, Algorithms, Probabilistic and Experimental Methodologies. First International Symposium, ESCAPE 2007, Hangzhou, China, April 7–9, 2007. Lecture Notes in Computer Science. Vol. 4614. pp. 1–11. doi:10.1007/978-3-540-74450-4_1. ISBN 978-3-540-74449-8. <a href="978-3-540-74449-8" target="_blank">978-3-540-74449-8</a> <a href="#fnref:5" class="footnote-back-ref">↩</a></li>
<li id="fn:6">Dósa, György (2007). "The Tight Bound of First Fit Decreasing Bin-Packing Algorithm is FFD(I) ≤ 11/9OPT(I) + 6/9". In Chen Bo; Mike Paterson; Zhang Guochuan (eds.). Combinatorics, Algorithms, Probabilistic and Experimental Methodologies. First International Symposium, ESCAPE 2007, Hangzhou, China, April 7–9, 2007. Lecture Notes in Computer Science. Vol. 4614. pp. 1–11. doi:10.1007/978-3-540-74450-4_1. ISBN 978-3-540-74449-8. <a href="978-3-540-74449-8" target="_blank">978-3-540-74449-8</a> <a href="#fnref:6" class="footnote-back-ref">↩</a></li>
<li id="fn:7">Coffman, E. G; Garey, M. R; Johnson, D. S (1987-12-01). "Bin packing with divisible item sizes". Journal of Complexity. 3 (4): 406–428. doi:10.1016/0885-064X(87)90009-4. ISSN 0885-064X. <a href="https://www.sciencedirect.com/science/article/pii/0885064X87900094" target="_blank">https://www.sciencedirect.com/science/article/pii/0885064X87900094</a> <a href="#fnref:7" class="footnote-back-ref">↩</a></li>
<li id="fn:8">Coffman, E. G. Jr.; Garey, M. R.; Johnson, D. S. (1978-02-01). "An Application of Bin-Packing to Multiprocessor Scheduling". SIAM Journal on Computing. 7 (1): 1–17. doi:10.1137/0207001. ISSN 0097-5397. <a href="https://epubs.siam.org/doi/abs/10.1137/0207001" target="_blank">https://epubs.siam.org/doi/abs/10.1137/0207001</a> <a href="#fnref:8" class="footnote-back-ref">↩</a></li>
<li id="fn:9">Coffman, E. G. Jr.; Garey, M. R.; Johnson, D. S. (1978-02-01). "An Application of Bin-Packing to Multiprocessor Scheduling". SIAM Journal on Computing. 7 (1): 1–17. doi:10.1137/0207001. ISSN 0097-5397. <a href="https://epubs.siam.org/doi/abs/10.1137/0207001" target="_blank">https://epubs.siam.org/doi/abs/10.1137/0207001</a> <a href="#fnref:9" class="footnote-back-ref">↩</a></li>
<li id="fn:10">Huang, Xin; Lu, Pinyan (2021-07-18). "An Algorithmic Framework for Approximating Maximin Share Allocation of Chores". Proceedings of the 22nd ACM Conference on Economics and Computation. EC '21. New York, NY, USA: Association for Computing Machinery. pp. 630–631. arXiv:1907.04505. doi:10.1145/3465456.3467555. ISBN 978-1-4503-8554-1. S2CID 195874333. <a href="978-1-4503-8554-1" target="_blank">978-1-4503-8554-1</a> <a href="#fnref:10" class="footnote-back-ref">↩</a></li>
<li id="fn:11">Johnson, David S; Garey, Michael R (October 1985). "A 7160 theorem for bin packing". Journal of Complexity. 1 (1): 65–106. doi:10.1016/0885-064X(85)90022-6. <a href="https://doi.org/10.1016%2F0885-064X%2885%2990022-6" target="_blank">https://doi.org/10.1016%2F0885-064X%2885%2990022-6</a> <a href="#fnref:11" class="footnote-back-ref">↩</a></li>
<li id="fn:12">Johnson, David S; Garey, Michael R (October 1985). "A 7160 theorem for bin packing". Journal of Complexity. 1 (1): 65–106. doi:10.1016/0885-064X(85)90022-6. <a href="https://doi.org/10.1016%2F0885-064X%2885%2990022-6" target="_blank">https://doi.org/10.1016%2F0885-064X%2885%2990022-6</a> <a href="#fnref:12" class="footnote-back-ref">↩</a></li>
<li id="fn:13">Yue, Minyi; Zhang, Lei (July 1995). "A simple proof of the inequality MFFD(L) ≤ 71/60 OPT(L) + 1,L for the MFFD bin-packing algorithm". Acta Mathematicae Applicatae Sinica. 11 (3): 318–330. doi:10.1007/BF02011198. S2CID 118263129. <a href="/wiki/Doi_(identifier)" target="_blank">/wiki/Doi_(identifier)</a> <a href="#fnref:13" class="footnote-back-ref">↩</a></li>
</ol>

First-fit-decreasing bin packing open-in-new

First-fit-decreasing bin packing