This paper considers a Train Shunting problem which occurs in cargo train organizations: We have a locomotive travelling along a track segment and a collection of n cars, where each car has a source and a target. Whenever the train passes the source of a car, it needs to be added to the train, and on the target, the respective car needs to be removed. Any such operation at the end of the train incurs low shunting cost, but adding or removing truly in the interior requires a more complex shunting operation and thus yields high cost. The objective is to schedule the adding and removal of cars as to minimize the total cost. This problem can also be seen as a relaxed version of the well-known List Update problem, which may be of independent interest. We derive polynomial time algorithms for Train Shunting by reducing this problem to finding independent sets in bipartite graphs. This allows us to treat several variants of the problem in a generic way. Specifically, we obtain an algorithm with running time O(n^{5/2}) for the uniform case, where all low costs and all high costs are identical, respectively. Furthermore, for the non-uniform case we have running time of O(n^3). Both versions translate to a symmetric variant, where it is also allowed to add and remove cars at the front of the train at low cost. In addition, we formulate a dynamic program with running time O(n^4), which exploits the special structure of the graph. Although the running time is worse, it allows us to solve many extensions, e.g., prize-collection, economies of scale, and dependencies between consecutive stations.

We consider the discrepancy problem of coloring n intervals with k colors such that at each point on the line, the maximal difference between the number of intervals of any two colors is minimal. Somewhat surprisingly, a coloring with maximal difference at most one always exists. Furthermore, we give an algorithm with running time O(n log n + kn log k) for its construction. This is in particular interesting because many known results for discrepancy problems are non-constructive. This problem naturally models a load balancing scenario, where $n$~tasks with given start- and endtimes have to be distributed among $k$~servers. Our results imply that this can be done ideally balanced. When generalizing to $d$-dimensional boxes (instead of intervals), a solution with difference at most one is not always possible. We show that for any d >= 2 and any k >= 2 it is NP-complete to decide if such a solution exists, which implies also NP-hardness of the respective minimization problem. In an online scenario, where intervals arrive over time and the color has to be decided upon arrival, the maximal difference in the size of color classes can become arbitrarily high for any online algorithm.

We consider the classical problem of scheduling preemptible jobs, that arrive over time, on identical parallel machines. The goal is to minimize the total completion time of the jobs. In standard scheduling notation of Graham et al. [5], this problem is denoted P |rj, pmtn| Σj cj. A popular algorithm called SRPT, which always schedules the unfinished jobs with shortest remaining processing time, is known to be 2-competitive, see Phillips et al. [13, 14]. This is also the best known competitive ratio for any online algorithm. However, it is conjectured that the competitive ratio of SRPT is significantly less than 2. Even breaking the barrier of 2 is considered a significant step towards the final answer of this classical online problem. We improve on this open problem by showing that SRPT is 1.86-competitive. This result is obtained using the following method, which might be of general interest: We define two dependent random variables that sum up to the difference between the cost of an SRPT schedule and the cost of an optimal schedule. Then we bound the sum of the expected values of these random variables with respect to the cost of the optimal schedule, yielding the claimed competitiveness. Furthermore, we show a lower bound of 21/19 for SRPT, improving on the previously best known 12/11 due to Lu et al. [11].

This paper investigates the problem of scheduling jobs on multiple speed-scaled processors without migration, i.e., we have constant α > 1 such that running a processor at speed s results in energy consumption s#945; per time unit. We consider the general case where each job has a monotonously increasing cost function that penalizes delay. This includes the so far considered cases of deadlines and flow time. For any type of delay cost functions, we obtain the following results: Any β-approximation algorithm for a single processor yields a randomized βBα-approximation algorithm for multiple processors, where Bα is the αth Bell number, that is, the number of partitions of a set of size α. Analogously, we show that any β-competitive online algorithm for a single processor yields a βBα-competitive online algorithm for multiple processors. Finally, we show that any β-approximation algorithm for multiple processors with migration yields a deterministic βBα-approximation algorithm for multiple processors without migration. These facts improve several approximation ratios and lead to new results. For instance, we obtain the first constant factor online and offline approximation algorithm for multiple processors without migration for arbitrary release times, deadlines, and job sizes. All algorithms are based on the surprising fact that we can remove migration with a blowup of Bα in expectation.

Memory management is a fundamental problem in computer architecture and operating systems. We consider a two-level memory system with fast, but small cache and slow, but large main memory. The underlying theoretical problem is known as the paging problem. A sequence of requests to pages has to be served by making each requested page available in the cache. A paging strategy replaces pages in the cache with requested ones. The aim is to minimize the number of page faults that occur whenever a requested page is not in the cache.Experience shows that the paging strategy LEAST-RECENTLY-USED (LRU) usually achieves a factor around 2 to 3 compared to the optimum number of faults. This contrasts the theoretical worst case, in which this factor can be as large as the cache size k.One difficulty in analyzing the paging problem was the lack of an appropriate lower bound for the minimum number of page faults. We address this issue and propose a general lower bound which provides insight into the global structure of a given request sequence. In addition, we derive a characterization for the number of faults incurred by LRU.We give a theoretical explanation why LRU performs well in practice. We classify the set of all request sequences according to certain parameters and prove a bound on the competitive ratio of LRU, which depends on them. This bound varies between 2 and k, i.e., it includes the worst-case, but explains for which sequences LRU achieves constant competitive ratio. The classification is motivated from the structure of request sequences of practical applications: locality of reference and characteristic data access patterns. We argue that this structure yields values around 2 for our bound. Indeed, it is between 2 and 5 in extensive practical experiments.Furthermore, we study the paging problem with variable cache size, which was already considered previously. We show that this approach is not appropriate to explain the usual good performance of LRU. We measure the performance of LRU with the expected competitive ratio E[ALG]/E[OPT] and the expected performance ratio E[ALG]/E[OPT] in a diffuse adversary model and compare both measures. Our analysis yields that the expected competitive ratio gives a misleading answer.