[ index ]
|
Microsoft has flip-flopped on its scalability message, now touting "scale-up" rather than "scale-out" improvements. "If you look at what Microsoft said in 1997 at its Scalability Day event it was to scale out. They said, 'Don't use SMP (Symmetric Multiprocessors) -- applications will hit a knee in the curve, and you have to pay more to scale up.'
"Now they're back to SMP. Get out your disco ball and leisure suit--Microsoft has taken you back to 1985," Shimp quipped.
Oracle, with its roots in Unix databases, has offered 64-bit support for years.
| Parellel Processing (PP) | vs | Symmetric Multiprocessing (SMP) |
| PP splits the problem domain into pieces and assigns each to its own processor with dedicated memory | SMP splits the range among available processors with shared memory. | |
| PP runs a copy of the solver on each processor and requires message passing at internal boundaries using multiprocessing languages like PVM and MPI. | SMP uses the multithreading paradigm and requires "index independence," i.e., no dependencies between different ranges | |
| PP is more scalable (two to thousands of processors) | SMP less scalable (two to tens of processors) because it is limited by shared memory bandwidth. | |
| PP requires significantly more code modifications. | SMP requires fewer code modifications. |
2003-04-23 15:33:42
©2009 Echo3 Online Services, LLC