I don't think the language necessarily helps or hinders the solving the oversubscription problem. C++ and Java will have similar challenges .
As an application developer you typically try to find the "right" number of threads to allocate to get the best performance. TBB creates a pool of threads on initialization (typically) based on what it calculates to be the best number to fully utilize the processors without creating contention.
As you mentioned, when mixed with another framework that "right" number for the application might not be quite right since the other framework might also allocate its own threads. If you can know ahead of time that the other frameworks threads will typically be used then you can probably account for that when you initialize TBB.
I think what Mr. Reed alludes to is a more general problem that occurs when more and more applications become thread enabled (or you try to run many instances of your newly created threaded app). Now all of the applications are trying to use all of the processors all of the time. What is the "right" number of threads to create on startup in this case?
Quoting -
Robert Reed (Intel)
Well the eventual solution is to treat threads and their communications mechanisms as a common resource, shared by all applications and the libraries they use, much like Intel has promoted in collaborating with other OpenMP vendors to provide a common OpenMP library to be shared regardless of which OpenMP front end is being used, which is dependent on the compiler used. Microsoft has suggested some directions they are considering for their next Visual Studio release and Intel is paying close attention. Maybe someday the Qts and TBBs of the world will all be in one big happy family, sharing common thread pools and cooperating in a fashion unimaginable today. Or maybe we'll just muddle about like we have for the last twenty years ;-).
For what it's worth, Apple has also proposed a model for sharing threading resources across applications in their upcoming OS X release. It would be interesting to hear Intel's position on this framework and whether TBB might one day integrate with it or perhaps support something similar on other platforms.
http://images.apple.com/macosx/technology/docs/GrandCentral_TB_brief_20090608.pdf
"Grand Central Dispatch is a revolutionary, pervasive approach to multicore processing. GCD shifts the responsibility for managing threads and their execution from applications to the operating system. Mac OS X Snow Leopard provides APIs for GCD throughout the system, and uses a highly scalable and efficient runtime mechanism to process work from applications. As a result, programmers can write less code to deal with concurrent operations in their applications, and the system can perform more efficiently."
I think we'll see a lot of interesting progress in the next few years in the state of the art for developing multi threaded applications.
