Kerneels comments:
>Google APPS are FREE for all educational institution and I just finished
>an assignment for an Educational institution but the "old world" can not
>compete with FREE.

They are "free" provided the institution and its users agree to Google's
terms of service. To oversimplify only slightly, Google provides "free"
services in exchange for their ability to do just about anything they like
with your data. And they do.

Is that "free" deal a good deal? "It depends."

The Anne & Lynn Wheeler ID writes:
>note that tpc benchmarks include total cost numbers per operation ...

TPC benchmarks include dollar figures *advertised* as total cost per
operation, yes. They are not in fact total costs. They are very, very
partial. For example, they completely omit the biggest IT cost of all in
many/most countries: staffing/labor. Understandably so perhaps since it's
tough to write such benchmark standards, but costs that are merely hard to
calculate are no less real.

Moreover, a vendor is under no obligation to deliver the advertised TPC
configuration everywhere in perpetuity (or for much time at all). Also, if
you modify the configuration or contract terms in any way -- and you
inevitably must -- all bets are off.

No, I am not referring to IBM behaviors as far as I am aware. But there are
such vendor antics in the world. Is anyone really surprised?

Then of course there's the fact that no business actually runs a TPC
benchmark to fill orders, manage inventory, support a call center, run
payroll, forecast demand, and so on. While it might be possible to engineer
systems to a well-specified fixed benchmark -- to incorporate a "benchmark
mode" (yes, this happens) -- it's a lot harder to engineer well for
arbitrary business workloads. TPC also, so far as I am aware, does not
include any quality-related metrics. For example, as long as the system
doesn't overheat within the benchmark run time it'll score. (And I'm not
joking here. Many other processor cores have "turbo" modes which cannot
sustain continuous operations due to thermal limits in those modes. It's
very tough for benchmarks to keep up with this sort of stuff.)

Anyway, while benchmarks are "interesting," let's not put *too* much stock
in them. If anything they're getting progressively less interesting over
time as real-world workload demands get more complex. We've had quite a bit
of discussion in this forum about how difficult it is to have reliable,
deterministic measurements with things like out-of-order execution, cache
hits and misses (at multiple levels of cache), and so on. Those issues
don't even begin to scratch the surface of how complex this stuff has
become.

--------------------------------------------------------------------------------------------------------
Timothy Sipples
GMU VCT Architect Executive (Based in Singapore)
E-Mail: [email protected]
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to