the numbers that might be seen now on the Intel machines or the CM-5, of 25 microseconds to send a message from one side to another, to something that's on the order of tens of nanoseconds or hundreds of nanoseconds, in terms of overhead. And that's the big step. And I think everything else that we're doing here is to reduce overhead. No, it's not part of the operating system, but the operating system and the compiler for the operating system can use those techniques as well. Nothing I want to sneer at. S M I T H : Just a little quick note. Mach has given up on, I think partially but, privileged threads as well. P A P A D O P O U L O S : Well, I know. Yeah, I mean, clearly, the threads always have seen their abuses there; I didn't want to... S M I T H : Right out of the kernel... PAPADOPOULOS: support now. But you don't get much up? Kevin has a question. T H E O B A L D : About Anant's comment equating VLSI placement algorithms with thread scheduling for compilers: would you elaborate on
/lp/association-for-computing-machinery/programming-compilation-and-resource-management-issues-for-p3xNoj3Ybf