Citation Context

...se most of the problems caused by machine failures, slow machines, and networking hiccups are dealt with automatically by the MapReduce library without operator intervention. Furthermore, it is easy to improve the performance of the indexing process by adding new machines to the indexing cluster. 7 Related Work Many systems have provided restricted programming models and used the restrictions to parallelize the computation automatically. For example, an associative function can be computed over all prefixes of an N element array in log N time on N processors using parallel prefix computations [6, 9, 13]. MapReduce can be considered a simplification and distillation of some of these models based on our experience with large real-world computations. More significantly, we provide a fault-tolerant implementation that scales to thousands of processors. In contrast, most of the parallel processing systems have only been implemented on smaller scales and leave the details of handling machine failures to the programmer. Bulk Synchronous Programming [17] and some MPI primitives [11] provide higher-level abstractions that make it easier for programmers to write parallel programs. A key difference bet...

Citation Context

... based on the observation that the techniques used for tuning code for the memory hierarchy are similar to those for developing parallel algorithms. Other primitive parallel operations: The scan-model=-=[6]-=- is an EREW PRAM model extended with unittime scan operations (data independent prefix operations), i.e., it assumes that certain scan operations can be executed as fast as parallel memory references....

Citation Context

...and locality, they are also efficient units for failure recovery. 6 Related Work MapReduce is similar to models that employ scan primitives or parallel prefix schemes to express parallel computations =-=[17, 4]-=-. Dubey has recently suggested the use of similar primitives in order to easily parallelize and schedule recognition, mining, and synthesis computations [9]. Concepts similar to MapReduce have also be...

Citation Context

...tion, some form of reduction, perhaps using only a fixed set of operators, and later scans (parallel prefixes) and permutation operations. In approximately chronological order, these models are: scan =-=[32]-=-, multiprefix [170], paralations [100, 171], the C* data--parallel language [111, 165], the scan-vector model and NESL [33--38], and CamlFlight [109]. As for other data-parallel languages, these model...

Citation Context

... the time taken from the point at which the current thread calls the barrier routine until the time this single thread is resumed. The barrier synchronization library routine is implemented in a scan =-=[3]-=- style. For an n processor machine, O(n log 2 n) messages are sent, n per wave. The pattern formed by the messages is that of a butterfly network mapped onto a 3-d grid, with the messages sent each wa...

Citation Context

... tree (expected O(lgn)). as is usually the case, the application of the subparts involves nested parallelism. Table 2 shows several examples. As an example, consider a parallel variation of quicksort =-=[2]-=- (see Figure 3). When applied to a vector s, this version splits the values into three subsets (the elements lesser, equal and greater than the pivot) and calls itself recursively on the lesser and gr...

Citation Context

...e scan operation on asynchronous machines. Blelloch suggested that certain scan operations be included in the P-RAM model as primitives and shows how this affects the complexity of various algorithms =-=[1]-=-. The line-of-sight and radix-sort algorithms are discussed by Blelloch [2, 3]. The parallel solution of recurrence problems was first discussed by Karp, Miller and Winograd [9], and parallel algorith...