Rapid once deployed too

[Fairlight]

At Thu, May 04, 2006 at 12:28:37PM -0500 or thereabouts, 
suspect Mike Schwartz (PC Support) was observed uttering:
> 
>      Then the customer upgraded their 5-year old server to a dual Xeon box
> with 4 gigs of RAM, and now the filePro web transactions run about 100 times
> as fast as they did on the old box.  

No doubt that a XEON will actually run things much faster, but it's the
bandying about of numbers like this that starts me wondering.  100 times
faster?  So if it takes one second now, it was taking 100 seconds (1:40) on
the old machine?  Why am I finding that improbable?  For that matter, even
if it took .4 seconds on the new machine, it took 40 seconds on the old
one?  In a stateless environment like a web server?  What's the average
concurrent request count?  Because anything that scales by a factor of 100
is seeming a bit out of whack to me.

If one assumes that's correct, -and- one assumes that MHz is actually a
valid benchmark (it's not, especially considering bus speeds changed,
I/O throughput changed, cache sizes and throughput changed, memory rates
changed), we're talking about having to be running a 24MHz CPU as the old
system, give or take, if you just divide a raw 2400 by 100.  Even
accounting for the above performance updates, the old machine should have
had to been something like a PPro-200 or so to see that kind of increase.
And that's older than five years--more like eight.

I just have a problem with "about 100 times faster" without hard timings.
I'm sure it's much faster.  You can compile on these things and sometimes
it feels like you're catting a file at 14.4kbps rather than actually
compiling.  They're incredible.  But things want real input numbers to make
a valid comparison that includes a final number.  Otherwise I could only
really accept "a lot faster".

>     The box might be "loafing" down to a 0.01 workload most of the time, but
> there are still instantaneous bursts where even this NEW computer isn't as
> fast as we would like it. 

What is "as fast as you would like it"?  I've had someone say that a
2-second request for something wasn't fast enough, which is really ironic
when you consider that no matter how big a pipe or fast a system you have
on the client side (I have a P4-3GHz/2GB sitting on 1.5mbit DSL), PayPal's
home page often loads slower than 2 seconds.  I just tried it.  3 seconds,
and it was feeling speedy today. ebay took 5s.  Which is why I start
questioning it when people say that 2s is not fast enough for users, as by
and large the overall audience on the net is getting slower service than
that from many popular commercial sites.  Heck, the cnet.com page takes 5
seconds just to load everything, and I would argue they don't even actually
-provide- much of anything compared to PayPal. :) That's just to get the
ugly-as-heck front page out the door and onto the screen.  We're talking
"hitting enter to Done" in the browser here.  Latest update to Mozilla that
came out 2 nights ago.  Even Microsoft comes in at just over a 2-count
for their home page, and that's doing no real back-end processing that's
apparent--just serving up static data.

>     The really depressing part is that the cost of the new box was probably
> a LOT less than the time I billed out to them on this issue over the past
> several months...

It's not unexpected.  Time is usually the most expensive aspect to fixing a
problem, mostly because it's not a commodity item.  You can't mass produce
knowledge and training and sell it in volume at steep cuts.  There's no
discounting that due to the economics of supply and demand.  It takes what
it takes on an individual basis.

mark->