Clock speed is only a small piece of the picture. Other major factors
are how many clock cycles are required to execute an instruction, how
many bits or bytes can be manipulated by a single instruction, how many
instructions can concurrently be in the execution pipeline, how many
independent ports there are to memory, the width in bits/bytes of the
memory and I/O bus paths, etc.
Even 30 years ago, I/O channels on all but the smallest mainframes were
essentially independent processors in their own right with independent
memory access. The number of independent channels and aggregate I/O
transfer rate on a z/10 today can easily outpace a PC by several orders
of magnitude and the I/O data transfers don't degrade the general
purpose CP's effective speed.
When PCs were promoting the 16-bit to 32-bit architecture revolution a
number of years back, larger mainframes had already been exploiting
internal memory bus widths of 128-bits or more for over a decade.
Typically the mainframes performance has been optimized for high data
transfer rates, manipulating multi-byte data values (larger integer and
floating point values, data records) and multi-tasking, and a single
instruction may do some very complex data manipulations that would
require many PC instructions.
The PC architecture has historically been more concerned with optimizing
manipulation of individual bits and bytes, a capability indispensable
for image processing and driving pretty graphical user display
interfaces. And, before you can talk about doing useful
application-level data manipulation, that graphical display interface on
the typical PC can suck up a significant portion of the available CPU
power and available memory. Consumption of additional PC CPU resources
is imposed by limitations of MS Operating systems, for such things as
Anti-Virus protection applications which are unnecessary in a mainframe
environment.
A mainframe, even one with a slower clock speed, can typically run
circles around a PC when dealing with a large number of independent
users or with applications that are I/O intensive. On the other hand, a
PC of comparable clock speed can do a single CPU intensive task where
individual bit or byte manipulations predominate in less real time than
the typical mainframe.
JC Ewing
On 03/06/2010 07:44 PM, Lindy Mayfield wrote:
> This has been a very interesting thread for me. If I remember correctly from
> the time I saw the z/10 with plexiglass outsides and a hardware guy there to
> explain what was what, and one of the things he told me was the cpu speed was
> (IIRC) 4.77GHz.
>
> My laptop has a dual core 1.99GHz.
>
> You already see where I'm going with this. How does a z/10 get so much more
> done? Forgive me for my math and analytical skills, but seems like 4 laptops
> could equal the speed of a 1 cpu z/10. There is a huge chunk of this
> equation that I'm totally missing. How does a z/10 get so much more done?
>
> kind regards, Lindy
>
> -----Original Message-----
> From: IBM Mainframe Discussion List [mailto:[email protected]] On Behalf
> Of R.S.
> Sent: 7. maaliskuuta 2010 2:23
> To: [email protected]
> Subject: Re: z9 / z10 instruction speed(s)
>
> Edward Jaffe pisze:
> [...]
>> People with PC-only experience are always astonished when I tell them
>> about modern mainframe provisioning capabilities. They always assume
>> when your hard drive fills up you need a new one or when your CPU is too
>> slow you need a new one. What we do seems like magic to them.
>
> Yes, mainframe capabilities are excellent in this area. From the other
> hand they solve problems which exist only in mainframe world: CPU power
> adjustment. CPU shortage is bad thing on any platform, but mainframe is
> the only one where too much MIPS is not good. Why to downgrade a PC?
> The same apply to "specialty" processors.
>
--
Joel C. Ewing, Fort Smith, AR [email protected]
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
