<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
<i><font color="#000000">"Worse, increased complexity of each block
reclaims increased complexity of connectivity between blocks,"</font></i><br>
<br>
I don't understand this sentence. Reclaims?<br>
<br>
<i>Unfortunately, there are a number of sentences starting with
“unfortunately” that<br>
curb the enthusiasm around increased computational density. For once,
connectivity is<br>
not only local—it forms a hierarchy [12]: closely connected components
formunits that<br>
must connect to other units, forming larger units. In turn, the larger
units also connect<br>
to other larger units forming even larger functional blocks, etc.
Connectivity-wise, such<br>
larger blocks remain “far away” from each other. Worse, increased
complexity of each<br>
block reclaims increased complexity of connectivity between blocks,
which is achieved<br>
by reducing the thickness of wires and the distance between them. That
means an increase<br>
of resistance, capacity, and crosstalk. Resistance and capacity worsen
propagation<br>
speed in the wire. Crosstalk is the propensity of the signal in one
wire to propagate<br>
to a nearby wire by (in this case) electromagnetic field. At high
frequencies, a wire is just<br>
an antenna and crosstalk becomes so unbearable that serial
communication increasingly<br>
replaces parallel communication (a somewhat counterintuitive phenomenon
visible<br>
at all scales—USB replaced the parallel port, SATA replaced PATA as the
disk data<br>
connector, and serial buses are replacing parallel buses in memory
subsystems, all because<br>
of crosstalk. Where are the days when parallel was fast and serial was
slow?)<br>
<br>
</i>Interesting, but I think it's a diversion from the point. I'd just
delete it.<br>
<br>
<br>
<i>To make us sweat even more, heat issues also join the fray.
Increased density means<br>
smaller transistors and consequently lower power draw per transistor,
but it alsomeans<br>
more transistors within the same space need to be fed with power,
leading to an overall<br>
trend towards increased power draw and consequently heat emission. The
silicon substrate<br>
sustaining the circuitry needs to dissipate ever larger amounts of
heat—another<br>
trend that scales the wrong way.<br>
In related, late-breaking news, the speed of light has obstinately
decided to stay constant<br>
(immutable if youwish) at about 300,000,000 meters per second. The
speed of light<br>
in silicon oxide (relevant to signal propagation inside today’s chips)
is about two thirds<br>
that, and the speed we can achieve today for transmitting actual data
is significantly below<br>
that theoretical limit. That spells more trouble for global
interconnectivity at high<br>
frequencies. If we wanted to build a 10 GHz chip, under ideal
conditions it would take<br>
two cycles just to transport a bit across a 4-centimeter wide chip
while essentially performing<br>
no computation.</i><br>
<br>
Interesting but also not relevant.<br>
<br>
<i>To illustrate the rapid changes in today’s concurrency world and
also the heavy in-<br>
fluence of data sharing on languages’ approach to concurrency, consider
the following<br>
piece of advice given in the first edition of the excellent book</i><br>
<br>
I'd just say "the 2001 edition" and then omit "<i>The book was
published in 2001</i>.". Later on, you mention the second edition,
saying the 200x edition instead will put things in better context.<br>
<br>
<i>If you do want to define data shared across processes, you must
qualify its type with shared.</i><br>
<br>
To define data as being shared across threads, it must be qualified
with shared. (Note threads, not processes.)<br>
<br>
<i>in D, perThread has one allocated copy per thread.</i><br>
<br>
in D, perThread has a separate copy for each thread.<br>
</body>
</html>