[HARDWARE] Rc5 and an ALR board
enojon at delphi.com
Wed Sep 15 20:37:13 EDT 1999
Addendum to morning reply..
The win95 co-processor in the PM8500 is running at about 112kkeys/sec
@ 100mhzP5 on 7" PCI PC-card; the PPC side (MacOS) is humming @
374kkeys/sec. The effect has been that W95 grabs one 8*2^28 packet
while PPC has one 8*2^28 packet. So, I suppose summing the two
sets yields 486kkeys/sec +/- for i/o latency [all i/o goes through MacOS].
I should probably install an Orange Mic. card and ramp up to WinNT/OS2
on K6-III and do the G3/G4 ramp-up. I wonder what that would yield?
Let's see... 1130kkeys/sec + 2930kkeys/sec ...? But, it would exceed
stoney at sequent.com wrote:
> I wasn't able to fine the multiprocessor paper at anatech.com, but
> IBM has a simple but good paper contrasting multiprocessor and
> uniprocessor systems.
> Good Reading list
> Pfister, Gregory, "In Search of Clusters: The ongoing battle in lowly
> parallel computing" 2nd Ed, Prentice Hall 1998
> (Great intro to many of those gaps in our knowledge. Enables you to
> then read texts and understand them :)
> Beck et Al, "Linux Kernel Internals" 2nd Ed, Addison Wesley 1998
> Chapter 10 Multi Processing
> Stalling, William, "Operating Systems Internals and Design Prinicipals"
> 3rd Ed, Prentice Hall 1998
> Nichols, Bradford et Al, "Pthreads Programming", O'Reilly 1996
> On Tue, 14 Sep 1999 ENOJON at delphi.com wrote:
> > <<
> > Linux SMP kernels before 2.0 used a master slave lock and approached
> > the performance of a uniprocessor when too many IO requests (MMIO)
> > were made. Only recently has Linux had course grained locking.
> > A single application needs to be threaded (user threads) before you can
> > take advantage of multiple processors; otherwise you only gain the ability
> > to concurrently do other OS/utility tasks. If your application is split
> > into different processes then you don't need to be threaded. >>
> > Agreed. Multiple processing is only useful for threaded apps or
> > parallel computing apps where collision of memory, i/o & cache
> > access can be limited. I think anatech.com has a paper on the
> > concepts of SMP and fallacies of assuming 2 400mhz PIII's=800mhzPIII.
> > Interesting read.
> > But, the Rc5 has good threading--as far as I can tell from my OS/2
> > client running 4 threads in a single session on a uniprocessor.
> > The win95 client also does multi-threading (as well as it can for
> > w95/98).
> > Taking the multiprocessor concept a tweak further--a PowerMac PPC(120mhz)
> > with PC-compatibility card sharing a "folder" of buff-in is humming along
> > at about 2 (1*2^28) blocks per hour--slow, but.. should give someone
> > an idea.
> > The PC side is running w95; mac os on the other side. [yes, i know, i
> > know, i should be running linux, but linux doesn't have pc-card support?]
> > --
> > To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net
> To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net
To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net
More information about the Hardware