Singl e proce ssor supercom puters have achi eved great speed s and have been pu shing
hardw are technol ogy to the physi cal limit of chip manu facturing. But soon this trend
wi ll com e to an end, becau se there are physi cal and arch itectura l bounds, which limit
the computati onal power that can be achi eved with a single processor system . In this
bo ok, we study advanc ed computer architect ures that utilize p arallelism via mul tiple
proce ssing units. Whi le parallel computin g, in the form o f internally linked
proce ssors, was the main form of parall elism, advanc es in com puter networ ks has
crea ted a new type of parallel ism in the form of networked auto nomo us computers.
Inst ead of puttin g everything in a single box and tightly coupl e proce ssors to
memor y, the Int ernet achieved a kind of para llelism by loosely connectin g ever y-
thi ng outs ide of the box. To get the mos t out of a compute r system with intern al
or exte rnal parallel ism, designer s and softw are develo pers mus t under stand the
int eraction between hardwar e and software p arts of the system. This is the reas on
we wrote thi s book. We wan t the reader to under stand the powe r and limit ations
of multipr ocess or syst ems. Our goal is to apprise the read er of both the benefic ial
No comments:
Post a Comment