[nylug-talk] Comparing RISC with CICS processor

[alex at pilosoft.com]

On Fri, 12 Jan 2007, Gary Mort wrote:

> Paul Robbins wrote:
> > OK, go maybe I am confused Gary. What decision are you trying to make and
> > how can I/we help?
> >   
> 
> 
> Right now I'm just trying to get an idea of what speed processor/memory
> they need to replace an old dual pentium system(which ironically has
> locked me out so I can't check the processor specs, I think it was a 1.6
> or 1.8 Pentium III with 2G of memory).
> 
> Every system on their list has loads of memory and hard drive space.
> So I'm really looking to see if there is any easy rule of thumb(ie don't
> use a 400Mhz RISC CPU as it's too slow, if the speed is at least equal
> to the current speed there should be no worries, etc).
Yes, it will be much slower. Even 16*400Mhz going to suck. (Although, it 
depends on what your application profile is. If it is very heavily 
parallelizable, it might not suck that bad. But that is unlikely).

In words of Seymour Cray: "If you were plowing a field, which would you 
rather use: Two strong oxen or 1024 chickens?"
> 
> I've found a number of sites that do comparisons, but they compare
> modern systems to modern systems(go figure, their advertising for
> computer sellers and they compare systems you can actually buy!  What
> are the odds? :-))
> 
> Naturally, I was hoping for some magic formula, but I expect the truth
> is it's going to be pretty much hit or miss and it really SHOULD be
> tested - but based on budget they don't have much of a choice.
www.spec.org as far as benchmarks go.

However, without knowing more about the exact system in question, it is 
impossible to predict anything. You didn't say whether the system is CPU 
bound, memory bound, memory-bandwidth-bound, disk-io bound, etc...

-alex