home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.sys.amiga.programmer:17560 comp.sys.amiga.hardware:21805
- Path: sparky!uunet!noc.near.net!bigboote.WPI.EDU!nntp!ahansfor
- From: ahansfor@bigwpi.wpi.edu (Andrew L. Hansford)
- Newsgroups: comp.sys.amiga.programmer,comp.sys.amiga.hardware
- Subject: Re: CISC and RISC
- Date: 19 Dec 92 03:05:42
- Organization: Worcester Polytechnic Institute, Worcester, MA 01609-2280
- Lines: 46
- Distribution: world
- Message-ID: <AHANSFOR.92Dec19030542@bigwpi.wpi.edu>
- References: <70436@cup.portal.com> <amipb.04wr@amipb.gna.org> <37844@cbmvax.commodore.com>
- <AHANSFOR.92Dec17115522@bigwpi.wpi.edu> <1gsorsINN327@crcnis1.unl.edu>
- NNTP-Posting-Host: bigwpi.wpi.edu
- In-reply-to: tbills@cse.unl.edu's message of 18 Dec 1992 14:57:00 GMT
-
- >>>>> On 18 Dec 1992 14:57:00 GMT, tbills@cse.unl.edu (Trent Bills) said:
-
- tbills> In article <AHANSFOR.92Dec17115522@bigwpi.wpi.edu>, ahansfor@bigwpi.wpi.edu (Andrew L. Hansford) writes:
-
- tbills> |> True. But like exception handling, more support for this seems to go
- tbills> |> into CISC machines. As for multiple processors, you still incur the
- tbills> |> same cost for a context switch. Now you have more processors
- tbills> |> switching. Esp with microkernel architectures that instead of one
- tbills> |> switch to the kernel and back to the calling function (2 switches),
- tbills> |> you get a context switch to the kernel, a message is copied or
- tbills> |> remapped, a context switch to a user lever server, and then the server
- tbills> |> enters the kernel to pass a message back to the calling proces. Now
- tbills> |> we have twice and many context switches.
-
- tbills> The idea behind having multiple processors is that you only have a few
- tbills> active processes at any one time and that each processor can be more
- tbills> or less dedicated to that process.
-
- How many processors do you plan to have running? How many processes
- do you intend to have running? I bet that at most any time
- #processes > #processors.
-
-
- tbills> The cost of a context switch in a RISC machine isn't usually that much
- tbills> more that a CISC machine. The only real difference is that the CISC
- tbills> machine only has to execute one or two instructions to perform the
- tbills> context switch whereas the RISC machine will have to execute much more.
- tbills> However, every instruction in the RISC machine will only take one clock
- tbills> cycle whereas the instructions for a task switch in a CISC machine
- tbills> will take a rather large number of clock cycles. Every access to memory
- tbills> will take at least one clock cycle regardless of the type of processor.
- tbills> The result is larger code on the RISC machine, but roughly identical
- tbills> performance for both.
-
- Yes. The time that each takes is roughly the same. The RISC chips
- come out slightly faster. Still, the more registers the longer a
- context switch. And RISC machines, having the virtue of being of new
- design, have larger register files. The effect is that application
- performance has seen a greater increase in performance than the OS.
- Note that the OS functions on RISC still run faster than on CISC but
- not by as large of margin.
-
-
- -- Andrew Hansford
- ahansfor@wpi.wpi.edu
-
-