Approximate parallel scheduling. Part I: the basic technique with applications to optimal parallel list ranking in logarithmic time
read more
Citations
Parallel algorithms for shared-memory machines
Sorting in linear time
Models of machines and computation for mapping in multicomputers
Towards a theory of nearly constant time parallel algorithms
Planar separators and parallel polygon triangulation
Frequently Asked Questions (16)
Q2. What are the main contributions of the vc-86a?
The main contributions of [CV-86a] were the deterministic coin tossing technique and a methodology for scheduling that used as few reschedulings as possible.
Q3. What is the main contribution of this paper?
One of the main contributions of this paper is to provide an algorithm for performing approximate rescheduling deterministically in 0(1) time.
Q4. How do the authors solve the task scheduling problem?
The authors remark that the task scheduling problem will be solved by redistributing the tasks,while the processor scheduling problem appears to require the redistribution of processors.
Q5. How do the authors ensure that all the trees are complete and distinct?
To ensure all the trees are complete and distinct the authors need to partition T into complete subtrees (recall that T is the smallest tree associated with the collection; thus the authors are guaranteed that if the authors create distinct sized complete binary trees from T then no two trees in the collection will have the same size and the authors get a proper set of complete binary trees).
Q6. What is the way to solve a problem in parallel?
Given an input of size n the parallel algorithm employs a reducing procedure to produce a smaller instance of the same problem (of size ^ n/2, say).
Q7. What is the proof of Brent's theorem?
Any synchronous parallel algorithm of time t that consists of a total of x elementary operations can be implemented by p processors in time [x/p] + t.Proof of Brent's theorem.
Q8. What is the upper bound on the parallel time achievable using p processors?
the best upper bound on the parallel time achievable using p processors, without improving the sequential result, is of the form 0(Seq(n)/p).
Q9. What is the Euler tour technique on trees?
The Euler tour technique on trees, which is given in [TV-85,Vi-85], consists ofreducing a variety of tree functions into list ranking.
Q10. What is the main contribution of the SV-86a?
Part 2 of this research (the paper [CV-86c]) shows how to apply the approximatescheduling method together with the new list ranking algorithm in order to derive improved PRAM upper bounds for a variety of problems on graphs, including: connectivity, biconnectivity, minimum spanning tree, Euler tour and st-numbering.
Q11. What is the definition of an expander graph?
Definition: A bipartite graph G = (Vi,V2, E), with |Vi| = IVjl, is a (d, e, —^)-expander graph if for any subset U C Vj, with |U | ^ e IVjl, the set N(U) of neighbors ofvertices in U has size |N(U) | S: ( ) |U |, and G has vertex degree d.
Q12. How many wtCSj neighbors are in the useful large sets?
since the size of the collections in each useful large set S; lie in the range [2', 2'"'"^), the authors conclude that the giving collections in Sj, counting multiplicities, have weight at least 1/4 wtCSj).
Q13. How many leaves in the largest tree of objects in Cj?
More precisely, let 3 be the number of leaves in the largest tree of objects in Cj (3 = 2 '); if 3 ^ 8d then the number of objects transferred is 3/8d; otherwise, no objects are moved.
Q14. how to schedule tasks on an EREW PRAM?
The problem is to schedule the n tasks on an EREW PRAM of n/log n processors so that the tasks are completed in 0(log n) time; it is solved in Section 4.
Q15. How can the authors build an expander graph?
Further such an expander graph can be built in 0(log |Vi|) time using |Vi|/log |Vi| processors, for each fixed e, as the authors show in the following remark.
Q16. How many nodes are marked for removal simultaneously?
There must be a chain of length x ^ log n + 1 nodes which ends at (and includes) u and satisfies the following: 1. The last x— 1 nodes of this chain are marked for removal simultaneously.