Technical Editor:
Martin Paprzycki Dept. of Computer Science and Statistics Univ. of Southern Mississippi Southern Station 1506 Hattierburg, MS 39406-1506
[email protected]
Structured Development of Parallel Programs By Susanna Pelagatti
248 pager $44.95 Taylor & Francis London 1997 0-7484-0759-6
88
Swuctiwcd Deueloi,nrcnr ofPmdlc1 l’ropa7m presents a structured pmgranmiing methodolo w for parallel computations that ensures portability, 1)rng~~i’i~*,’alhiliiy, and good perfor~iiaiicc.Thc book‘s iiltimatr goal is to dcvclop a suitable progranuning language for p”rallc1 p g r a m m i n g and its compilcr. l’his language is meant to deliver typical parallel constriicu: (slmlctons)and their realinations (templates) on various arcliitccmrcs. T h c honk‘s first half presents a critical analysis uf the state of thc art nf parallel software clcvclopineni. It also closely exaniincs scvcral existing approaches to parallcl prngramming, concluding that template-based systems a m thc Ixst cmnpromise. In this approach, the prugraminer selects skeletiins a n d their cotivcrsinn rules, then iises tbem tn huild a prngrarn. Its perforinancc mightnnt match that of a lowlevel graph-based approach, hut it is predictable and casily cnsiires prograinnuihility and portability. T h e ho&‘s second half descrilics thc P3L tcmplatc-based tuethodology a n d its realization as thc P3L language and its compilcr, nffering application cxanqAcs. T h e author maintains that thc icmplate-based system gives rise to accurate performance rnodcls for the skeletons library dcsigncr as well as for the progratnnmr. T h e technical and mapping &tails arc left to the skeleton library &signer, who can folly cxploit specific propertics o l p”rticu1ar skclctnns. T h e 1’31, mctliodolngy incurporates a stiuill set of hasic skeletons and their cnmhinatiun rules. Skeleton selection is based on the analysis of existing approaches. T h c skclctons rcflect typical constructs t h a t parallel program designers usc.
‘l‘hc P3L methndology might be a good starting point for developing efficient higlilevel laiigoagcs for parallel prngrainming. It suggcsts how tu ensure compromise hctwccn p e r f ~ ~ m ~ a an icdepomhiliiy and programtnaIrility. In any case, we should not trcat it as soincthing clnsed and finally establisliedhigh-level parallel programming languages continue to clcvclop and imprnve. Such high-level languages would lct the programiner conccntrate less o n the details of the machine’s architectore and tnorc o n the algorithm’s dcsign. ’l‘hc lack nf high-level langilagcs is otic of the major h t a c l e s hainpering large, complex softwwc projects and the dcvclopmcnt nf computational algorithms. Currently, the progress or these Ianguages is scvcrcly delayed compared to tlic pirc parallel hardware performancc. An efficient, high-level language for parallel progratnming available on computers with parallel processors and nn clusters of machines used for distrihotcd computatinns woold be an important tool fhr people developing gcnera1 theoretical a n d al’plication-orieiited algorithms. ’l‘his hookshould interest people working on parallel algorithms, hut, more importantly, it should interest researchers and safiware engineers developing languages for parallel computatinns. It might also be of interest to 1)utlt undergraduatc and graduate computer science siurlcnts hccause it does lint rcquirc any special hackground. It can supplement material fnr cnurses devoted to programming languages and compilation techoiqoes, cspccially fur high-level parallcl programming. IEEE Concurrencv
It might look a hit strange-publishing reviews of books published so inany years ago (PracticalPnrnllci Progmmming, 1995), or published more recently, but containing rather old material (PnrflilelAlgorith~h~, published in 1997, but conraining material from 1993-94). Inthesase ofthc first hook, [asked for the review after talkiligwithProfessorYalamov. H e thinks it’s a good textbook and uses it in his parallel progrpmming course, which he teaches a t the University of Rousse. Many ofus (myself included) are constantly in search of good textbooks (they call be a hit old as lotig as they’re still useful), which is why the seconcl book, it i s really a pity that it took the AMs three years to publish it-especially because ress p,?rallel computing in die context of combinatorial and graph problems. This relative lackof cove to request this review. -Marcin Pizprzycki, “Rook Reuiews”Ediror
Parallel Algorithms ~
~~
Parallel Algorithms Sandeep N. Bhatt. ed. 162 pages
$45.00 American Mathematical Society Providence, R.I 1997 0-8218-0447-2
~~
~
October-December 1999
Parallcl computers have solvcd many largescale priihlems, typically those arising frorn scientific and cnginrering applications. IIuwever, massive parallclism for nonnurncrical applications has received less actcnti
