When the market for parallel computers finally took off in the early 90s, it was dominated by moderately sized SMP servers. Their simple model of uniformly shared resources, such as CPUs, memories and I/Os, were appealing features compared with the MPP of the time.
Today, technology pressure and the quest for scalability have pushed SMP implementations far away from the simple model of the 90s. There are strong indications that this trend will continue.
I will present the state-of-the art in SMP technology and show how increased bandwidth demands have changed the implementation over time. I will also present some simple examples of possible optimizations to applications, runtime system and architecture that can overcome some of the current SMP bottlenecks. Finally, I will give my prediction of the SMP future.