Someone from Intel Labs came and gave a talk
here a few weeks ago, and dropped an interesting
fact -- they've learned how to distribute a processor
team within a single time zone or two pretty well
(say, Oregon, Santa Clara, and Folsom), but the
amount of daily interaction needed for a custom
chip makes distributing a single design between,
say, India, Oregon, and Israel not easy at all. So,
processor design jobs are stickier to a region,
for the same reason full-custom VLSI is so hard
in general (and avoided whenever possible) --
breaking the design apart horizontally (architecture,
logic, circuit, layout) and vertically (ALU, register files,
caches) leaves everyone with a schedule full of meetings
each week to make sure details aren't falling through the
cracks. The only practical way to outsource is to create
the whole team in a region, and finding 200 specialists
to fill all the roles a processor needs takes a generation of
preparation (successful example: Intel Israel).
> Stop making things that have been done 1000x > before unless a) It's for fun/educational purposes. > b) You're going to do something someone else hasn't.
Sometimes, it takes a lot of almost-ready's to get to the one design that is ready for mass acceptance. Some examples:
-- Hyper-text was a mature idea by the time TBL took his shot at it with HTML/HTTP. But he put the known concepts together in a nice way with some new ideas, and it (eventually) took off.
-- C was not a language with oodles of brand new ideas, compared with the languages that preceded it. But again, it put known concepts together in nice way with some new ideas, and it took off.
My point is, the old saw of "doing the same thing over and over again, while expecting a new result, is a sign of insanity" doesn't apply to design. Because, as the French say, in great matters, no detail is small. A few changes in the details can turn a "yet another" into the "next big thing".
Happy memories of my 2340A, but I dunno if I would recommend spending cash to buy one on the used market... they ran so hot, and their mean-time-to-failure reflected it. We had a few dozen under our control (research lab + class lab), and there would always be a few with little yellow Post-Its on them waiting to be sent out for repair.
If you only own a PowerBook, and you develop software that is cross-platform for OS X and Linux, it's very convenient to be able to run both Linux and OS X on the same machine for testing.
> What you're talking about has happened. > And failed.
Sometimes it takes a lot of tries for a new thing to stick. There were many hypertext systems before Tim Berners-Lee did his variant. There were many pen computers before Palm broke through. Early failures do not indicate a hopeless idea.
The coolest seminars I've seen on campus this semester have been virologists and immunologists making real-time movies of cells under attack (virology) and pre-empting attack (immunology). To sit there in the audience and watch a movie of a flu virus (tagged with a flourescent marker to look red) tricking its way into a cell, maneurvering to the nucleus, and attacking it, is just stunning. And the immunologists have the same sort of movies with dendritic cells dancing with antigens. Yes, I realize its a long way from having the movie to understanding the science behind the movie sufficiently to reach the clinic, but that fact doesn't make it any less stunning...
See this EE Times article on a
partnership between Matrix Semiconductor (3D
write-once ROM technology, spun off of Stanford)
and Nintendo. Also note the large number of
"flash cards" included for the price (4), and a
description of using the cards that make it sound
like a "write-once" technology.
Apple wants to sell clusters to the scientific, creative, and academic market -- witness Xserve, and the decision to put Virginia Tech at the front of the queue for 1100 G5s. Buying RedHat gives Apple a (Linux) customer base in those machine rooms -- just like buying Shake gave them a (Linux) customer base in CG. And just like Apple Shake supports Linux, but makes it financially advantagous to move to OS X, RedHat-the-Apple-subsidiary could do a similiar migration strategy, underpricing service this time to win OS X market share instead of application software as in the Shake case.
Note that in the analyst conference call last week, Apple CFO Fred Anderson noted that part of the reason Apple keeps a 4B+ cash reserve is to be ready for a large aquisition that "changes the rules of the game"...
The US can learn a lot by understanding
the Italian model of success. Using the
conventional measures of the cost of
doing business in a country, Italy does
not look so good. But yet, many Italian
companies do a thriving export business,
with at least product design happening in the
country. The espresso machine in your
Starbucks was designed in Italy, Vespa
scooters, clothing, etc.
What all these products have in common
is that they embody Italian culture in
their design, in a way people notice,
and in a way that would not survive
outsouring the design elsewhere. Good
corporate branding plays a role, but
the mega-brand of Italy -- romance,
style, talent -- is what closes the
sale.
The IETF midcom group has been working
on solutions for passing media streams through
NATs and other middleboxes for a few years now.
One protocol, STUN, is already a standards-track RFC, and the
group has other tools in progress. These tools
work with the IETF multimedia suite (SDP, SIP,
RTP, etc).
> Few people buy stock based on their opinion > of the ethics, or lack thereof, of a company.
VCSIX (the Vanguard Index Fund for the Calvert Social Index) has $150 M in assets. Given that's mostly in mom-and-pop retirement portfolios, that's a non-trivial number of folks putting an ethical screen on their investments... and it only represents a fraction of the socially-conscious mutual fund world.
In some contexts, the size of the file sent to the printer is an important consideration. Coding a page as the shortest computer program that can generate the page is "the best you can do". Of course, whether or not dvips is generating the optimal program is another issue entirely.
The number of people who can buy a Mountain View (Los Altos, Palo Alto, etc) house cash exceeds the number of houses on the market. The semi-retired gentry likes the lifestyle of the area, and 30 years of tech success has produced a lot of gentry. And in many many cases, these folks can design prototypes with their own hands, and think its fun to do so. I don't think a comeback hinges on re-locates -- it hinges on lifers.
In music, I think we have good case studies now which show that it is possible to "say all that is worth saying" within a genre. Look at the "big band music" genre -- by the end of WWII it had all been said, and the innovators moved on to create new types of jazz. The bands that play that music today do it as historical preservation. Given a set of instruments, and stylistic rules for writing to the instruments, there is only so much one can say.
One lasting contribution Inktomi made was validating Networks of Workstations in a commercial context. Remember, at the time they started, the chief competition was (DEC-era) AltaVista, which used the search engine as an example application for multi-way SMP boxes. Today, you don't see >2-way SMP used in massive deployments of applications that are easy to parallelize, but back when Inktomi started NoW's were novel.
Miyamoto should try to tackle the problem of making a game that is a compelling musical instrument, simpler to learn than conventional instruments, but which gives people the same feeling of personal expression. This would span the range of kids and adults, be naturally multi-player, and take the console in a new direction. And he has the muscle-pixel-sound intuition to know what this game should be.
2003 could be an interesting year if Sun bought RedHat. The numbers work: at 862M market cap, RHAT is 10% of Sun's market cap. Plus, Sun has 5B+ of cash in the bank.
It may be worth it, just for the transient effect of having all of the hardware vendors who partner with Redhat scurry to find a new Linux vendor. That's six months of FUD, during which Sun can launch their new Linux hardware as a safe choice against Dell and IBM and HP. The long-term value of RedHat is an extra.
Music folks would have said the same thing in 1984, and they would have been right. As technology advances, production values become easier to create -- give it enough time and it will happen for video too.
RTP scales upwards OK in bandwidth -- there are uncompressed SMPTE packetizations for HDTV in progress, which is a fair number of bits moving through the pipe. The advantage of designing to RTP (as an industry, if not for your current prototype project) is that an IF stream is a media stream, and so a lot of the associated tools RTP offers are a perfect fit (session management tools, security tools, etc). So, the standards writing job in minimized to just the basics needed to describe the radio-specific stuff.
> It would be nice to not have to do arp or > deal with any real buffering, but just build > the frame as you go, and stuff > the CRC at the end. Send the frame as > a broadcast packet.
I was thinking about going one (very thin) layer up -- and really making UDP IP packets, unicast or multicast, in those GB Etherframes, and have those UDP packets use RTP. This way, IF radio becomes "just another RTP media type", along with all the audio and video codecs. If you define the RTP packetization the right way, all sorts of radio hardware and applications could conform to it. Yes, it would be the most bandwidth-intensive RTP packetization to date, however, the SMPTE HDTV uncompressed packetization that's underway in AVT now is moving a fair number of bits, so 12-bit quantization of a 20Mhz IF is not too far out of the mainstream.
Consider defining a general-purpose RTP packetization for IF radio, and running through the IETF. This would let hardware folks innovate to a known stack: Gig etherframes, UDP, RTP w/ your packetization. You could also spec out a SIP framework, so the hardware can start up its own sessions.
Slashdot Mirror
Someone from Intel Labs came and gave a talk here a few weeks ago, and dropped an interesting fact -- they've learned how to distribute a processor team within a single time zone or two pretty well (say, Oregon, Santa Clara, and Folsom), but the amount of daily interaction needed for a custom chip makes distributing a single design between, say, India, Oregon, and Israel not easy at all. So, processor design jobs are stickier to a region, for the same reason full-custom VLSI is so hard in general (and avoided whenever possible) -- breaking the design apart horizontally (architecture, logic, circuit, layout) and vertically (ALU, register files, caches) leaves everyone with a schedule full of meetings each week to make sure details aren't falling through the cracks. The only practical way to outsource is to create the whole team in a region, and finding 200 specialists to fill all the roles a processor needs takes a generation of preparation (successful example: Intel Israel).
> Stop making things that have been done 1000x
> before unless a) It's for fun/educational purposes.
> b) You're going to do something someone else hasn't.
Sometimes, it takes a lot of almost-ready's to get
to the one design that is ready for mass acceptance.
Some examples:
-- Hyper-text was a mature idea by the time TBL
took his shot at it with HTML/HTTP. But he put
the known concepts together in a nice way with
some new ideas, and it (eventually) took off.
-- C was not a language with oodles of brand new
ideas, compared with the languages that preceded
it. But again, it put known concepts together in
nice way with some new ideas, and it took off.
My point is, the old saw of "doing the same thing
over and over again, while expecting a new result,
is a sign of insanity" doesn't apply to design.
Because, as the French say, in great matters,
no detail is small. A few changes in the details
can turn a "yet another" into the "next big thing".
Happy memories of my 2340A, but I dunno if I ... they ran so hot, and their
would recommend spending cash to buy one on
the used market
mean-time-to-failure reflected it. We had a few
dozen under our control (research lab + class lab),
and there would always be a few with little yellow
Post-Its on them waiting to be sent out for repair.
If you only own a PowerBook, and you
develop software that is cross-platform
for OS X and Linux, it's very convenient
to be able to run both Linux and OS X
on the same machine for testing.
Cisco sells wireless VoIP handsets.
> What you're talking about has happened.
> And failed.
Sometimes it takes a lot of tries for a
new thing to stick. There were many
hypertext systems before Tim
Berners-Lee did his variant. There
were many pen computers before
Palm broke through. Early failures do
not indicate a hopeless idea.
The coolest seminars I've seen on campus this ...
semester have been virologists and immunologists
making real-time movies of cells under attack
(virology) and pre-empting attack (immunology).
To sit there in the audience and watch a movie
of a flu virus (tagged with a flourescent marker
to look red) tricking its way into a cell, maneurvering
to the nucleus, and attacking it, is just stunning.
And the immunologists have the same sort of
movies with dendritic cells dancing with antigens.
Yes, I realize its a long way from having the movie
to understanding the science behind the movie
sufficiently to reach the clinic, but that fact
doesn't make it any less stunning
See this EE Times article on a partnership between Matrix Semiconductor (3D write-once ROM technology, spun off of Stanford) and Nintendo. Also note the large number of "flash cards" included for the price (4), and a description of using the cards that make it sound like a "write-once" technology.
Apple wants to sell clusters to the scientific, creative,
...
and academic market -- witness Xserve, and the
decision to put Virginia Tech at the front of the
queue for 1100 G5s. Buying RedHat gives Apple a
(Linux) customer base in those machine rooms -- just
like buying Shake gave them a (Linux) customer
base in CG. And just like Apple Shake supports Linux,
but makes it financially advantagous to move to
OS X, RedHat-the-Apple-subsidiary could do
a similiar migration strategy, underpricing service
this time to win OS X market share instead of
application software as in the Shake case.
Note that in the analyst conference call last week,
Apple CFO Fred Anderson noted that part of the
reason Apple keeps a 4B+ cash reserve is to be
ready for a large aquisition that "changes the
rules of the game"
What all these products have in common is that they embody Italian culture in their design, in a way people notice, and in a way that would not survive outsouring the design elsewhere. Good corporate branding plays a role, but the mega-brand of Italy -- romance, style, talent -- is what closes the sale.
The honor code system at Caltech worked ... it had a
well at minimizing cheating
lot of structural support, though.
The IETF midcom group has been working on solutions for passing media streams through NATs and other middleboxes for a few years now. One protocol, STUN, is already a standards-track RFC, and the group has other tools in progress. These tools work with the IETF multimedia suite (SDP, SIP, RTP, etc).
> Few people buy stock based on their opinion
... and it only represents a fraction
> of the ethics, or lack thereof, of a company.
VCSIX (the Vanguard Index Fund for the
Calvert Social Index) has $150 M in assets.
Given that's mostly in mom-and-pop retirement
portfolios, that's a non-trivial number of
folks putting an ethical screen on their
investments
of the socially-conscious mutual fund world.
See chart
In some contexts, the size of the file sent
to the printer is an important consideration.
Coding a page as the shortest computer
program that can generate the page is "the
best you can do". Of course, whether or not
dvips is generating the optimal program is
another issue entirely.
The number of people who can buy a Mountain
View (Los Altos, Palo Alto, etc) house cash
exceeds the number of houses on the market.
The semi-retired gentry likes the lifestyle
of the area, and 30 years of tech success
has produced a lot of gentry. And in many
many cases, these folks can design prototypes
with their own hands, and think its fun to
do so. I don't think a comeback hinges on
re-locates -- it hinges on lifers.
In music, I think we have good case studies
now which show that it is possible to "say all
that is worth saying" within a genre. Look
at the "big band music" genre -- by the end
of WWII it had all been said, and the innovators
moved on to create new types of jazz. The
bands that play that music today do it as
historical preservation. Given a set of
instruments, and stylistic rules for
writing to the instruments, there is only
so much one can say.
One lasting contribution Inktomi made
was validating Networks of Workstations
in a commercial context. Remember, at the
time they started, the chief competition
was (DEC-era) AltaVista, which used
the search engine as an example application
for multi-way SMP boxes. Today, you don't
see >2-way SMP used in massive deployments
of applications that are easy to parallelize,
but back when Inktomi started NoW's were novel.
Miyamoto should try to tackle the problem
of making a game that is a compelling musical
instrument, simpler to learn than conventional
instruments, but which gives people the same
feeling of personal expression. This would span
the range of kids and adults, be naturally
multi-player, and take the console in a new
direction. And he has the muscle-pixel-sound
intuition to know what this game should be.
2003 could be an interesting year if Sun
bought RedHat. The numbers work: at 862M
market cap, RHAT is 10% of Sun's market
cap. Plus, Sun has 5B+ of cash in the bank.
It may be worth it, just for the transient
effect of having all of the hardware vendors
who partner with Redhat scurry to find a new
Linux vendor. That's six months of FUD, during
which Sun can launch their new Linux hardware
as a safe choice against Dell and IBM and HP.
The long-term value of RedHat is an extra.
Music folks would have said the same thing
in 1984, and they would have been right.
As technology advances, production values
become easier to create -- give it enough
time and it will happen for video too.
RTP is the real-time protocol -- it's how
l
media streams (video, audio, etc) get
sent around UDP networks in the IETF
view of the world. See:
http://www.ietf.org/html.charters/avt-charter.htm
RTP scales upwards OK in bandwidth --
there are uncompressed SMPTE packetizations
for HDTV in progress, which is a fair number
of bits moving through the pipe. The
advantage of designing to RTP (as an industry,
if not for your current prototype project) is
that an IF stream is a media stream, and so
a lot of the associated tools RTP offers are
a perfect fit (session management tools,
security tools, etc). So, the standards
writing job in minimized to just the basics
needed to describe the radio-specific stuff.
> It would be nice to not have to do arp or
> deal with any real buffering, but just build
> the frame as you go, and stuff
> the CRC at the end. Send the frame as
> a broadcast packet.
I was thinking about going one (very thin)
layer up -- and really making UDP IP packets,
unicast or multicast, in those GB Etherframes,
and have those UDP packets use RTP. This way,
IF radio becomes "just another RTP media type",
along with all the audio and video codecs. If
you define the RTP packetization the right way,
all sorts of radio hardware and applications
could conform to it. Yes, it would be the most
bandwidth-intensive RTP packetization to date,
however, the SMPTE HDTV uncompressed packetization
that's underway in AVT now is moving a fair
number of bits, so 12-bit quantization of a
20Mhz IF is not too far out of the mainstream.
> We've also thought about Gig ethernet.
Consider defining a general-purpose RTP
packetization for IF radio, and running
through the IETF. This would let hardware
folks innovate to a known stack: Gig
etherframes, UDP, RTP w/ your packetization.
You could also spec out a SIP framework,
so the hardware can start up its own sessions.
Bob did a lot of datacomm science behindo ral_histories/transcripts/lucky.html for details.
the major modem advances that came out
of Bell Labs. See http://www.ieee.org/organizations/history_center/