[SATLUG] question of the measurement of processing power past and present

Jennie Haywood jehaywood at gmail.com
Tue Aug 12 14:30:54 CDT 2008


On Tue, Aug 12, 2008 at 10:46 AM, Todd W. Bucy <toddwbucy at grandecom.net> wrote:
> On Tue, 2008-08-12 at 10:32 -0500, Todd W. Bucy wrote:
>> I am still hacking away at this paper.  The paper looks at F/OSS in
>> terms of a gift economy.  One of the trueism that has always been noted
>> about gift economies is that they occur in environments which are
>> relatively wealthy and where competition for resources is not as high;
>> the potlatch of the Kwakutial is often cited as such an example.  I am
>> making the basic argument that this need not be the case and am offering
>> the birth of hackers and the F/OSS community as such an example.  If we
>> consider that the production environemnt for computer software was
>> severly constricted in terms of resources (computing power) forced an
>> environment not dictated by competition for resources (as the tragedy of
>> the commons would predict) but was driven by cooperation in terms of
>> time sharing protocols.
>>
>> what I need to know is it fair, in terms of raw processing power, to
>> compare a measurement of 0.06 kips of the IBM 650 in 1954 to the
>> petraFLOPS generated by the IBM Roadrunner of today (or for that matter
>> the giga flops generated by the the intell q6600).  Obviously the newer
>> processors (and super computers) of today are much faster then they were
>> in the 50's and 60's.  I guess what I am looking for is something that
>> illustrates the dramatic increase of computing power over the last 50 or
>> so years.
> I am also concered with my change in unit of measure, from mips to flops


I'm not sure that's a change in units so much at different measures
for 'types' of calculation.  "flops" is for floating point operations
and generally only used for scientific/math calculations. Floating
point instructions are generally more involved to execute.    'mips'
is millions of instructions per second which starts to lose relevance
when you have 1000+ nodes.

If you look at some of the performance data for the IBM pSeries 590
(specifically for  Oracle or DB2 servers) you'll see 'transactions'
per second.  What's a transaction?    And even when you talk mflops on
the 590 you still have to think about how many processors were used?
The way the hardware had been set up if you had a 595 (the largest one
they were making) with 64 processors ( using the particular cards with
2 processors per card) to get the best performance ( for scientific
calulations)  you shut off half the processors, thus only using one
processor on each card.

The interesting thing with the newest IBM supercomputers is they are
running SUSE not AIX.


-- 
Jennie Haywood
----
Everyone is crazy. It's just a matter of degree.
--
The oak tree in your backyard is just a nut that held its ground.


More information about the SATLUG mailing list