This preview shows page 1. Sign up to view the full content.
Unformatted text preview: o me’ Is P aying Of f ,” New Yo rk Times, April 23, 2002. Multico re, massively parallel, and grid co mputing are all related in that each attempts to lash
to gether multiple co mputing devices so that they can wo rk to gether to so lve pro blems. Think o f
multico re chips as having several pro cesso rs in a single chip. Think o f massively parallel
superco mputers as having several chips in o ne co mputer, and think o f grid co mputing as using
existing co mputers to wo rk to gether o n a single task (essentially a co mputer made up o f multiple
co mputers). While these techno lo gies o f f er great pro mise, they’re all subject to the same
limitatio n: so f tware must be written to divide existing pro blems into smaller pieces that can be
handled by each co re, pro cesso r, o r co mputer, respectively. So me pro blems, such as simulatio ns,
are easy to split up, but f o r pro blems that are linear (where, f o r example, step two can’t be started
until the results f ro m step o ne are kno wn), the multiple-brain appro ach do esn’t o f f er much help. Massive clusters o f co mputers running so f tware that allo ws them to o perate as a unif ied service
also enable new service-based co mputing mo dels, such as sof t ware as a service (SaaS) and
cloud comput ing. In these mo dels, o rganizatio ns replace traditio nal so f tware and hardware that
they wo uld run in-ho use with services that are delivered o nline. Go o gle, Micro so f t, Salesf o rce.co m,
and Amazo n are amo ng the f irms that have sunk billio ns into these Mo o re’s Law–enabled server
f arms, creating entirely new businesses that pro mise to radically redraw the so f tware and
hardware landscape while bringing gargantuan co mputing po wer to the little guy. (See Chapter 10
"So f tware in Flux: P artly Clo udy and So metimes Free".) Mo ving ‘brains’ to the clo ud can help
increase calculating perf o rmance even when we can’t pack mo re pro cessing brawn into o ur devices,
but the clo ud requires a lo ng-distance co nnectio n that’s a lo t slo wer and perhaps less reliable than the quick ho p f ro m sto rage to pro cesso r that o ccurs inside mo st co nsumer electro nics. Mo o re’s Law will likely hit its physical limit in yo ur lif etime, but no o ne really kno ws if this
“Mo o re’s Wall” is a decade away o r mo re. What lies ahead is anyo ne’s guess. So me techno lo gies,
such as still-experimental quantum co mputing, co uld make co mputers that are mo re po werf ul than
all the wo rld’s co nventio nal co mputers co mbined. Think strategically—new waves o f inno vatio n
might so o n be sho uting “surf ’s up!” K E Y TAK E AWAYS
Most modern supercomputers use massive sets of microprocessors working in parallel.
The microprocessors used in most modern supercomputers are often the same commodity chips
that can be found in conventional PCs and servers.
Moore’s Law means that businesses as diverse as financial services firms, industri...
View Full Document
This document was uploaded on 01/31/2014.
- Winter '14