Re: RAM-CPU Singularity
- From: Kurt Van Etten <x@xxx>
- Date: Thu, 09 Feb 2006 10:04:47 -0500
yaoziyuan@xxxxxxxxx wrote:
There are already successful (maybe) examples of transparent tradeoffs
between different kinds of computing resources, such as virtual memory.
Well, this may be a somewhat crude example, but from my mainframe programming days I recall that the Syncsort sorting routine does just this sort of thing. If the dataset is small enough to fit in core memory, then one sort algorithm is used; if the working data needs to be written to disk then another algorithm is used; and if the working data is going to be written to tape then yet another algorithm is used. The program also tests I/O channels for their relative capacity before deciding how to allocate the temporary storage areas, etc.
But it seems to me that variations on this kind of "fast memory vs. slow memory" distinction are the only practical kind of tuning between resources that you can do. It's hard to imagine (for me, at least), how you could make much headway against a slow CPU. Sure, if you have more RAM that might help. But if extra memory makes a significant difference in program speed, you're going to want to use it whether you've got a slow CPU or a fast CPU. Maybe this idea might have some applicability to distributed computing, where you could spread your work out to a number of slow CPUs, and the optimal number of CPUs might depend on their memory capacity.
--
Kurt Van Etten
kurt (at) learning (concat) computation (dot) com
http://www.learningcomputation.com/blog/
.
- References:
- RAM-CPU Singularity
- From: yaoziyuan
- RAM-CPU Singularity
- Prev by Date: Re: RAM-CPU Singularity
- Next by Date: Call for Papers: IMECS 2006 (multiconference of 14 engineering & computer science conferences)
- Previous by thread: Re: RAM-CPU Singularity
- Next by thread: Sorting large number of records
- Index(es):
Relevant Pages
|