Algorithms Sequential & Parallel: A Unified Approach

On the surface, serial vs. parallel algorithms are about the newest trends in parallel processing, with their attendant problems of memory allocation, race conditions et al, and their upsides of speed and efficiency. You'd think that Cormen, the go to book on algorithms today, would give a LOT more coverage to parallel, as these authors do. Not the case! (Granted,  Introduction to Algorithms  is now 4 years old, and costs $80, not $300 like some texts. THIS book, however, can be found for $40 bucks or so US new, a very fair deal for the currency, relevance and quality of the text. You can find Cormen used for that price, but the information is dated and parallel very sparse).A level deeper, the parallel vs. serial issue is not just about hardware, it is about the problems themselves. Some problems, like extending pi at length, REQUIRE a serial approach, because as dynamic systems, their results feed back into the next step. On the other hand, finding gobs of primes for encryption can be "broken down" into parallel tasks, then combined to give faster results.I actually disagree with the "author reviewer" here and in the book that parallel is the wave of the future (mostly) because that's the wave of processing. The reason I take exception with that is that dynamical systems, with feedback algorithms resulting in convergence, divergence, chaos, oscillation, strange attraction, etc. ARE cutting edge, and cannot be "broken up" into parallel pieces, because feedback itself implies sequence. In fact, some of the hottest parallel research right now is about "handing" partial results back and forth, making the parallel construct mimic sequential in dynamic feedback situations, or even combine serial and parallel. Even a simple algorithm that guesses square roots by estimating high, then using successive feedback iterations to converge, is far more efficient than, say parallel sieves in other math functions that don't require feedback.This is NOT to say that this book is not of great value. It really is the ONLY text that pays this much attention to parallel, and their even more current related cousin, distributed algorithms. Because serial hardware allocation has been worked on for so long, sequential algorithms have fewer processor issues than parallel, so you will be spending a lot more time on the hardware side as you get into parallel. That's not to say that the sticky issues with memory in serial are "solved" -- brute force memory is both a cop out and a "good enough" solution.But compared to serial, parallel is a field needing a lot more contributors, so your study of this text will do your career a lot of benefit! Load balancing, the transfer overhead of passed messages, data locking of shared memory processing (requiring either additional bus cycles or partial serialization of your parallel algorithm), and many other sticky issues abound as OPPORUNITIES that require much more emphasis.This is promoted as of value for a half year rather than a full year course, but I think that's aiming low, and giving Cormen too much credit. This text really has a LOT of value vs. price not due to "page count" but price for currency as well as relevancy of information. Frankly, Cormen covers a lot of material not needed in today's environment. So if you're here for self study rather than "required reading" the choice is far simpler: GET THIS INSTEAD of Cormen! Most of us are on a budget today, and this text has far more relevant signal vs. noise than Cormen, regardless of course length. Even if parallel is cutting edge and fraught with issues, these authors do a GREAT job of continually relating them back to traditional serial processes. This is important, because the solutions generally COMBINE all three-- sequential, parallel and distributed, including, in some cases, WITHIN the algorithm.A final note to a select audience: I realize that a LOT of buyers today get these texts as interview prep. If that's the case, this is a volume you CANNOT MISS. For the price, BUY IT, because anyone interviewing today will give equal weight to your knowledge of parallel, NOT found in Cormen et al. I'm not trying to scare you or cost you money, it's just that I do interviews all the time as a manager at an algorithm design group and am continually shocked at how little even highly skilled graduate candidates know about this topic. Even scanning this will be worth your while-- RENT IT from Amazon if you must! I know you're just out of school, possibly broke, and looking for work (been there), so don't think I give this perspective lightly. If it were this or Cormen, I buy this even vs. Cormen used, but if your career is at stake, get Cormen used and this new, you should be able to do both for under $80, which is astonishing given the cost of texts in other related fields today. Parallel will distinguish you WAY MORE in the interview, as it is still relatively rare to hear any depth about it from newer grads.Library Picks reviews only for the benefit of Amazon shoppers and has nothing to do with Amazon, the authors, manufacturers or publishers of the items we review. We always buy the items we review for the sake of objectivity, and although we search for gems, are not shy about trashing an item if it's a waste of time or money for Amazon shoppers. If the reviewer identifies herself, her job or her field, it is only as a point of reference to help you gauge the background and any biases.

Excellent book

Amazon requires a rating for this review to be posted; otherwise, I wouldn't give a rating. I'm a coauthor, hence not unbiased. The real purpose of this review is to give you additional information. Much of what appears below is quoted from or paraphrased from my review of the previous edition.The dominant textbook in the field of computer algorithms is Introduction to Algorithms, by Cormen et al. This is a very fine book. However, we have written Algorithms Sequential and Parallel in a very different style, which we feel will give significant advantages to many who use our book. Points of difference between these texts include the following:1. Algorithms Sequential and Parallel has a unified approach to the presentation of sequential and parallel algorithms. Students of 21st Century computing will need to learn parallel algorithms, which are often closely related to their sequential analogs. Ours is the first algorithms text to integrate presentation of sequential and parallel algorithms so that readers can understand their relationships. This integrated treatment also frees the instructor from the common practice of spending most of an algorithms course on the study of sequential algorithms, with the last 2 or 3 weeks devoted to parallel algorithms - a practice that may mislead students into thinking that parallel computing is still the immature specialization of researchers, rather than a core technology for the 21st Century.2. Algorithms Sequential and Parallel does not compete with Cormen et al. in the scope of topics covered. The Cormen et al. text, whose 3rd edition is over 1300 pages, seems designed for at least a year sequence in the study of algorithms. By contrast, Algorithms Sequential and Parallel, 3rd edition, is just over 400 pages. This makes it suitable for a one-semester study of algorithms, appropriate for many undergraduate and first- or second-year graduate programs; also, it means a careful shopper can purchase Algorithms Sequential and Parallel at a lower price.Among the major topics of Algorithms Sequential and Parallel:* Mathematical tools used in the analysis of algorithms, including introductions to mathematical induction, recursion, and the Master Theorem.* A variety of sequential and parallel models of computation (including the RAM, PRAM, linear array, mesh, hypercube, pyramid, mesh-of-trees, and modern platforms ranging from coarse-grained models to networks of workstations, clusters, grids, and clouds).* Fundamental algorithms (including data movement operations such as broadcast, gather and scatter, permutation exchange, and sorting), semigroup computations, parallel prefix, searching, and their sequential and parallel implementations.* Paradigms such as recursive divide-and-conquer.* Algorithms for a variety of applications areas (matrix operations and a variety of other numerical problems, computational geometry, image processing, graph problems, etc).Each chapter concludes with exercises at varying levels of difficulty.Note that a disinterested reviewer gave the first edition of Algorithms Sequential and Parallel a rating of 5 stars (out of 5) in SIGACT News 34 #2, June, 2003, pp. 3-5.

Loved this book. A great starting point to develop a unified approach to combined sequential and parallel algorithmic solutions to different problems!

This is a math book. There are no practical examples and even if you love math you'll get bogged down following along with their proofs. If this book were an algorithm designed to get information into your brain it would be like... O(n).