Abstract:In dynamic load balancing, we wish to distribute balls into bins in an environment where both balls and bins can be added and removed. We want to minimize the maximum load of any bin but we also want to minimize the number of balls and bins affected when adding or removing a ball or a bin. We want a hashing-style solution where we given the ID of a ball can find its bin efficiently. We are given a balancing parameter $c=1+\epsilon$, where $\epsilon\in (0,1)$. With $n$ and $m$ the current numbers of balls and bins, we want no bin with load above $C=\lceil c n/m\rceil$, referred to as the capacity of the bins. We present a scheme where we can locate a ball checking $1+O(\log 1/\epsilon)$ bins in expectation. When inserting or deleting a ball, we expect to move $O(1/\epsilon)$ balls, and when inserting or deleting a bin, we expect to move $O(C/\epsilon)$ balls. Previous bounds were off by a factor $1/\epsilon$. These bounds are best possible when $C=O(1)$ but for larger $C$, we can do much better: Let $f=\epsilon C$ if $C\leq \log 1/\epsilon$, $f=\epsilon\sqrt{C}\cdot \sqrt{\log(1/(\epsilon\sqrt{C}))}$ if $\log 1/\epsilon\leq C<\tfrac{1}{2\epsilon^2}$, and $C=1$ if $C\geq \tfrac{1}{2\epsilon^2}$. We show that we expect to move $O(1/f)$ balls when inserting or deleting a ball, and $O(C/f)$ balls when inserting or deleting a bin. For the bounds with larger $C$, we first have to resolve a much simpler probabilistic problem. Place $n$ balls in $m$ bins of capacity $C$, one ball at the time. Each ball picks a uniformly random non-full bin. We show that in expectation and with high probability, the fraction of non-full bins is $\Theta(f)$. Then the expected number of bins that a new ball would have to visit to find one that is not full is $\Theta(1/f)$. As it turns out, we obtain the same complexity in our more complicated scheme where both balls and bins can be added and removed.

An asymptotically optimal algorithm for generating bin cardinalities

ON THE FLUCTUATION IN THE RANDOM ASSIGNMENT PROBLEM

Optimal on-line algorithms for variable-sized bin covering

An Asymptotic Competitive Scheme for Online Bin Packing.

The Asymptotic Distribution and Berry–Esseen Bound of a New Test for Independence in High Dimension with an Application to Stochastic Optimization

Load Balancing with Dynamic Set of Balls and Bins

Linear Time-Approximation Algorithms for Bin Packing

Parallel Balanced Allocations: The Heavily Loaded Case

The Probability to Hit Every Bin with a Linear Number of Balls

The power of thinning in balanced allocation

Expected number of uniformly distributed balls in a most loaded bin using placement with simple linear functions

Balanced Allocations: Caching and Packing, Twinning and Thinning

A Generalization of Multiple Choice Balls-into-Bins: Tight Bounds

Optimal Discrete Uniform Generation from Coin Flips, and Applications

LIL and the Approximation of Rectangular Sums of B-valued Random Variables when Extreme Terms Are Excluded

A Note On the Size of Largest Bins Using Placement With Linear Transformations

Long-term balanced allocation via thinning

A Dense Hierarchy of Sublinear Time Approximation Schemes for Bin Packing

An Asymptotic Lower Bound for Online Vector Bin Packing

Load balancing under $d$-thinning

Asymptotically Optimal Strategies For Combinatorial Semi-Bandits in Polynomial Time