|
Snow and ice this morning, but it will probably melt by this afternoon. It's
a good morning not to have to take out the car or go anywhere in particular.
Drivers around here drive bad enough in the best of conditions.
Last night I caught up on some reading (while the wife was on the Web
<g>), and found an interesting article in
Communications of
the
ACM
about the Netscape design philosophy. It put some
of the developments during the past 3 or so years into better perspective
and explains a few things about what went wrong for Netscape from about version
3. If you recall, the N share of the browser market peaked in early 1996
at about 90%, to fall below 50% only a year and a half later. (The
article refreshingly refrains from any accusations about Microsoft's role
in this demise, and focuses only on the design philosophy issue, namely
that...)
This development can to a substantial part be attributed to the early Netscape
commitment to strictly adhere to cross-platform development. The
marketing mantra became the promise to run on every platform,
and to release new versions more or less simultaneously for all of them.
A tall order.
In other words, ideally the application code (browser or server) was to have
been written completely independent of platform, resting on an abstracted
API layer (Netscape Portable Runtime -- NSPR) which would mediate system
calls for each OS. This position was summed up in 1997 as "The advantage
is that (the) product truly works cross-platform without rewriting. We are
able to release on all platforms and it pretty much works the same."
While a noble goal, the company soon experienced increasingly severe pressure
to retreat into progressively more OS-dedicated code in order to meet design
objectives within reasonable time and cost constraints.
In fact, the cross-platform components probably accounted for no more than
20% of the performance sensitive server code. Browser code, although initially
built far more cross-platform, showed a trend from perhaps 80% shared code
in v3 dropping to 60% with v5 Mozilla. There was also considerable resource
strain in supporting not only Windows and Mac versions, but also seven or
more different flavors of Unix on the server side. Additionally, NSPR abstraction
caused performance hits (typically factor 2), a critical issue for servers.
The cross-platform design mandated "lowest common denominator" interface,
which reflected badly in comparison with competitor products that quickly
used the latest GUI features of each OS. As it was, Netscape engineers put
considerable effort into finetuning code for performance on Wintel platforms.
This also explains why Netscape was initially so thrilled about Java, because
the Java promise was to realize true app-code independence of OS. The Java
Virtual Machine (VM) provided a third party and standard implementation of
what NSPR was trying to realize, and which at the time looked to attain universal
acceptance. Given the VM as an independent standard, Netscape could then
concentrate on the products instead of diverting significant resources to
developing and marketing their own version of a VM. When Sun's Java was found
inadequate to the demands, however, Netscape went in 1997 to developing an
in-house version. This was soon scrapped (in 1998, along with Mozilla 6
development) and coding retreated to C/C++.
Netscape failed to attain the full benefits of cross-platform design, at
least without paying far greater development penalties than envisioned. A
conservative in-house estimate figured the cross-platform approach penalty
as an at least 10-15% greater cost in time and cost as compared to the effort
of developing a dedicated OS version of the same product. While this would
normally be expected to reap significant advantages in the longer term, the
accelerating development of platform OS and web/GUI features pushed for
continually faster revisions in the abstraction layer -- the VM never stabilized.
Costs for in-house Java development and bottlenecks in all-platform testing
were also proving to be daunting.
It may be noted that while Microsoft had also dabbled in cross-platform "p-code"
development (for e.g. Word), this was soon abandoned due to the speed and
GUI penalties. For example, IE was originally coded specifically only for
Windows 95, and as other platform versions were released, these were built
on OS specific code bases each time.
Came so the AOL/Sun acquisition of Netscape after the Mozilla PD release.
Once again the focus is presumably on an all-Java browser, which begs the
question: where is the cross-platform support in the shape of a pervasive
and capable Java-VM? One may also wonder if the Sun involvement will force
optimization to Sun's Solaris at the expense of support for say Windows.
The more general issue of component reuse (write once, use many times) has
a long history of discussion in the Object Oriented field, where experience
frequently shows that the commendable theory of application is compromised
by realworld design issues that necessitate recurring, large and expensive
rewrites of code for specific circumstances. As it happens, this is not news
-- much the same is seen in the realm of physical engineering, and has been
true at least as long as humans have been constructing things (see for example
the many pyramids of Egypt that embody and preserve a fascinating record
of design evolution).
The Independent (UK online) has a
pretty
dismal
view
of the American (NATO) use of depleted uranium (DU) in Kosovo (and elsewhere).
This use was noted already in the Gulf War, if anyone remembers (and
I believe I once remarked about it somewhere in DisISay),
because DU in a shell or warhead is an effective tank killer (high mass
translates into deeper "burn-through" penetration of armor). Still, that
does sum up much of arms development throughout human history: throw
heavier things faster. The parallel paradigm is: more bang for the
buck.
I seem to recall a really old SF novel, probably late 50s or early
60s, that mentioned this as an efficient way of penetrating enemy spacecraft
meteor shields. Back then, DU and tactical theatre nukes were if not deployed
(in Europe) already, at least seen as the next logical step in standard
battlefield munitions.
The snow continues, mostly large and fluffy flakes, and the evening sets
in already before 4 p.m.
A draft report from Working Group Two of the Intergovernmental Panel
on Climate Change (IPCC), discusses among other things the effects of
as yet largely unexplored "threshold triggers" that may drive drastic global
climate changes.
Some more or less well known examples include the instability of the thermohaline
circulation that drives warmer water to the North Atlantic (aka Gulf Stream)
and the apparent collapse currently in progress of the west Antarctic ice
sheet.
A more recent instability concerns the release of methane -- a potent greenhouse
gas -- from frozen gas hydrates trapped in permafrost or in ocean floor sediment.
It is known that these hydrates store enormous volumes of the gas. (As
an aside, sudden releases of frozen hydrates underlies at least one theory
trying to explain the Bermuda Triangle disappearances.) Some recent
research indicates that a substantial melting occurred some 55 million years
ago, when the oceans warmed perhaps 5 degrees or so, releasing vast quantities
of gas. This in turn rapidly drove global climate into runaway "hot" mode,
further heating the oceans and releasing more gas. The current thought is
that eventually both carbon dioxide and the hydrocarbons precipitated out
of the atmosphere and got locked back into ocean and sediments.
While the scale of these events dwarf anything that humans have (yet) managed,
we must not get complacent, because the trigger mechanisms are by definition
(forced non-linear systems) extremely sensitive to small perturbations due
to the positive feedback loops. In cases such as this, one does not usually
notice any shift in overall equilibrium until it is far too late, because
another characteristic is that the shift to a new balance point proceeds
extremely rapidly once past the cusp.
The IPCC faces criticism from many who do not accept that climate change
is anything we need to worry about in the foreseeable future. Nor is it generally
accepted that the current scale of human activity has any significant impact
on these large scale climate processes. Luckily, however, research continues
apace and the models are becoming better -- or more interesting,
which is perhaps really the same thing.
Speaking of global warming, I just discovered that I had left the frying
pan on the hotplate at max. For. A. Long. Time...
It was an a gratuitous opportunity to explore the sessile effect of water
drops on extended and variously structured superheated surfaces... Oh well,
it's a solid cast iron pan, so after things cooled down, it ended up being
a matter of having burned out the old fat and able to rub in some new. Not
the first time, by either me or the wife.
|
Hi,
welcome to this week's journal. The update-link
(above) points to where I last added some text, which should simplify your
keeping up to speed. Of course, this assumes I remember to move it, and you
may still have to scroll back a bit and see if I've updated more than once
since you last visited, but that is easily done.
.
Open yours now!
