Microsoft trying to make sense of multicore

From a marketing perspective, multicore processors are an easy sell. Two brains are better than one. Four brains are better than two. You get the idea.

The challenge is that a whole lot of computer software has been designed to take advantage of ever-faster brains, not a computer packed full of them. It's a particular challenge for desktop and mobile computers. On the server and supercomputing side, the notion of parallel computing has been around for some time.

In the PC world, software makers have been scrambling to find new ways of thinking as Moore's law is quickly taking the chip world into a realm where there may be dozens or hundreds of processing units, or cores, on a single chip. In its latest attempt to figure out what to do with all those cores, Microsoft said Friday it is setting up a joint research center in Barcelona with the Barcelona Supercomputing Center.

The BSC-Microsoft research center "will focus on the way microprocessors and software for the mobile and desktop market segments will be designed and interact over the next 10 years and beyond," Microsoft said in a statement. "The advent of many- and multi-core processor computing architectures will make it possible to deliver enormous computational power on a single chip, with profound implications for the way software is developed."

The center will look at new approaches to software design.

"To optimize the designs and interactions of multicore processors and software, we need to start from parallel programming," Barcelona Supercomputing Center director Mateo Valero said in a statement. "The way to deal with this multicore architecture challenge is to bring together computer architects and programming language experts."

[eightwings]

Square Peg vs. Round Hole

It's a good thing for them that the leadership at Microsoft realizes that whoever cracks the parallel programming nut will dominate computing in this century. This is the latest in their efforts to bring sanity into the chaos of parallel multicore computing. Their recent hiring of super-computing expert Dan Reed is a case in point. However, I think that most parallel computer architects are trying to fit a square peg into a round hole. Here's a quote from a blog article on the matter:

"Adding more processing cores to a CPU should have been a relatively painless evolution of computer technology but it turned out to be a real pain in the ass, programming wise. Why? To understand the problem, we must go back to the very beginning of the computer age, close to a hundred and fifty years ago, when an Englishman named Charles Babbage designed the world?s first general purpose computer, the analytical engine."

Quoted from "Parellel Programming, Math and the Curse of the Algorithm:
http://rebelscience.blogspot.com/2007/10/parallel-programming-math-and-curse-of.html

[rcrusoe]

Assign applications to specific cores?

Maybe MS could work out a way to assign applications to
specific cores.

For example, they could assign antivirus and antispyware to one
core, firewalls and updates to another, etc.

That way we might have enough uninterrupted processing power
from our new PC to play Solitare and run WordPad at the same
time. ;)

[RompStar_420]

other OSed already do this

Why should I wait 10 years for Microsoft to figure this out ? I am already using an Operating System that does this very well, it's called OS X and Linux - get with the program.

Why would I wait 10 years for them to get or maybe not get it ? The ones that already get it, will also advance.

[The_Decider]

This is the future of end-user multi-processor computers

Adding a second core will never, ever, ever result in 2x execution speed. There are many reasons for this: memory syncing from the cache level down(this is an issue for compiler and OS writers, not application developers), context switching, many end user apps are not inherently parallel, etc.

A more significant improvement will be seen by running an application in one core, so the other cores can run different apps.

The usage and challenges of end-user computing is completely different from supercomputing and other MP systems and API's. Comparing them, as MS seems to want to will lead nowhere.

[Lolo Gecko]

maybe the key to MPP is ...

minimizing the number of CPUs, not maximizing them? if we could get the number of CPUs down to zero; e.g. discard the processing but keep the memory, MPP may become a reality. it is highly doubtful that going in the opposite direction by increasing the number of CPUs will ever yield a natural solution. maximizing memory throughput and eliminating central processing seems more intuitive. or, maybe I'm just another clueless gecko. or, maybe we can make parallel processing so minutely granular that it evolves from MPP into MPM (Massively Parallel Memory)

[kfarnham]

ParallelFX and Threading Building Blocks

It's surprising that this article doesn't mention ParallelFX, Microsoft's .NET multithreading library that was released as a community technology preview a few months ago. ParallelFX is somewhat similar to Threading Building Blocks, a C++ template library invented by Intel, then turned into an open source project: see http://ThreadingBuildingBlocks.org

Both ParallelFX and Threading Building Blocks offer paths to multithreading old and new applications without the developer having to manage the threading details. TBB (the more mature of the two) is gaining a following in universities around the world.