Paper
haha
IndustrialResearchDr.RoyLevin.pptxIs Corporate Computing Research Dead?
Roy Levin
April 11, 2019
Abstract: Over the past 50 years, corporate research labs have produced several of the most widely used and essential innovations in computing. Yet most of the once-great labs (and in some cases the companies that created them) are gone, leading to the conventional wisdom that corporate computing research is an idea whose time has come and gone. In this talk, I examine the past successes of corporate computing research, explore its purported demise, and discuss whether we can recreate the conditions that engendered so many foundational technologies of the software age.
1
The Argument
Computing research has made possible huge advantages in society.
Much of it was publicly funded (via government agencies).
Some of it came from corporate research labs.
I will argue that:
Many of the most significant and influential systems came from corporate labs.
Significant systems rarely come directly from other research organizations.
I will conclude that:
Society loses without corporate computing research labs.
But corporate research labs are disappearing. What can we do about it?
April 11, 2019
Is Corporate Computing Research Dead?
2
X
you!
On the next four slides, I will examine the four key words: “corporate”, “computing”, “research”, and “labs”.
2
What do I Mean by “Corporate”?
Research occurs in a profit-making setting.
In contrast to academia or government research labs
Risk and reward in a corporation
Corporations are evaluated (too) regularly by Wall Street.
Rewards for researchers (money, promotion) are based on the impact of work.
Judgments of impact occur on widely varying timescales.
Job security is (ultimately) based on impact.
April 11, 2019
Is Corporate Computing Research Dead?
3
In a corporation, everything is ultimately judged by its contribution to financial results. How and how often that judgment occurs, and when actions based on that judgment are taken, vary widely.
3
What do I Mean by “Computing”?
“Computer science is the study of the phenomena surrounding computers.”
— Allen Newell and Herbert Simon CACM, March, 1976 (Turing Award lecture)
Computing research isn’t limited to hardware and programming.
Phenomena arising from computing keep evolving and expanding.
April 11, 2019
Is Corporate Computing Research Dead?
4
The present “big data”/machine-learning revolution and the algorithmic work arising from collaborations in the biological sciences are just two examples illustrating the breadth of the field of computing.
4
What do I Mean by “Research”?
Basic
Novel
Relevant
Innovative
Strategic
Incremental
Theoretical
Ambitious
Timely
Applied
Derivative
Long-term
Useful
Wacky
Is Corporate Computing Research Dead?
5
April 11, 2019
The most influential labs do research with all of these properties!
A dictionary definition:
Investigation or experimentation aimed at the discovery and interpretation of facts, revision of accepted theories or laws in the light of new facts, or practical application of such new or revised theories or laws
ASK: Before I give you my view, I’d like to hear yours: What does the word “research” conjure up for you?
ASK: If someone described to you a piece of something they called “research”, how would you decide whether it was research or not?
Broad definition – in any given setting, it requires some qualifying adjective(s).
All of these adjectives can describe kinds of research, but some are mutually exclusive.
5
What do I Mean by “Lab”?
Zen and the Art of Research Management (Naughton & Taylor):
Hire only the very best people.
Trust them.
Protest your researchers from external interference.
Remember that you are a conductor, not a soloist.
[there’s more, but too much for today]
To learn more:
A Perspective on Research Management (Levin)
April 11, 2019
Is Corporate Computing Research Dead?
6
The most influential corporate research labs reflect a management philosophy pithily outlined in Naughton and Taylor’s paper. As of 2019, all two pages are visible at: https://link.springer.com/chapter/10.1007%2F0-387-21821-1_32
The four bullets here give a sense of that philosophy. It’s described in more detail in my paper. As of 2019, available as: https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/osrresearchmgmt.pdf
6
The Research Landscape (simplified)
Is Corporate Computing Research Dead?
7
Disruptive
Blue Sky
Sustaining
short-term
long-term
open-ended;
divergent
reactive;
convergent
Mission-focused
Portfolio
April 11, 2019
This is simplified because research doesn’t neatly fall into four disjoint buckets, but this picture does give a (simplistic) way to think about the goal, customers, and character of impact of research work based on the two indicated dimensions. Note that the boundaries between the quadrants are artificial – both axes are really continuous scales. The portfolio blob is merely intended to suggest that, whatever organization is represented, it has made some choices. And, of course, the portfolio represents a snapshot, since it changes over time.
7
Typical Academic Research Portfolio
Is Corporate Computing Research Dead?
8
April 11, 2019
Disruptive
Blue Sky
Sustaining
short-term
long-term
Mission-focused
Academic
Portfolio
open-ended;
divergent
reactive;
convergent
Academic work tends not to be mission-focused. (Exceptions: industry-funded, projects, DoD-type projects done semi-secretly). Longer-term work tends to be free of such problems. Funding agencies lean toward less blue-sky work now, so the lower-right is more populated and upper left goes only to the point that funding agencies don’t consider wacky. Nevertheless, upper right quadrant is still the focus, and much theoretical work falls here. Disruptive work tends to lead to start-up formation, which is (mostly) outside academia. Portfolio management isn’t really a concept in academia; to the extent that it happens, it is driven by the funding agencies.
8
Typical Government Lab Portfolio
Is Corporate Computing Research Dead?
9
April 11, 2019
Disruptive
Blue Sky
Sustaining
short-term
long-term
Mission-focused
Gov’t Lab
Portfolio
open-ended;
divergent
reactive;
convergent
Society can’t tolerate much disruption, so most of the work is in the lower half. A few blips into the upper half, especially in the blue sky region, represent particular programs (like quantum computing) whose potential payoff is so high that the risk is worth taking with the taxpayers’ money.
9
Typical Start-up’s Research Portfolio
Is Corporate Computing Research Dead?
10
April 11, 2019
Disruptive
Blue Sky
Sustaining
short-term
long-term
Mission-focused
Start-up
Portfolio
open-ended;
divergent
reactive;
convergent
Start-ups have to be strongly focused on their business, so all of their research investment – to the extent they have any at all – has to be in the lower part of the diagram, emphasizing growth of existing product line. As the company grows, the investment can push further into the “sustaining” area of the diagram.
10
Typical Corporate Research Portfolio
Is Corporate Computing Research Dead?
11
April 11, 2019
Disruptive
Blue Sky
Sustaining
short-term
long-term
Mission-focused
?
open-ended;
divergent
reactive;
convergent
Corporations vary widely in the way they manage their research portfolio and therefore in what that portfolio contains. There is no such thing as “typical”. Some have research labs, some seed researcher-types in advanced development groups. Among current companies, compare/contrast Microsoft, Facebook, Samsung, VMWare, Amazon, Google.
11
Balancing a Research Portfolio
April 11, 2019
Is Corporate Computing Research Dead?
12
Applied
Theoretical
Relevant
Strategic
Basic
Novel
Innovative
Incremental
Ambitious
Timely
Useful
Long-term
Wacky
Derivative
Here are those words from a few slides ago, all of which I said can characterize research at various times. It’s the job of the research manager to balance them appropriately based on the organization’s philosophy, mission, short- and long-term needs, budget, and other factors.
12
The 3 R’s of Balancing a Corporate Research Portfolio
Reward: Nature of pay-off
Who benefits?
How much?
When?
Risk: Likelihood and timing of payoff
Technical risk = can we create technology necessary for a project to work?
Organizational risk = can we get our results adopted?
Relationships:
to other research activities (dependence, overlap)
to the company’s business (relevance, competition)
Is Corporate Computing Research Dead?
13
April 11, 2019
How does a research manager balance all those dancing words on the previous slides? In considering a project for the portfolio, these are the questions that the manager has to answer.
Technical risk=how likely is it that we can actually create the necessary technology to make the project work.
Organizational risk=how likely is it that the fruits of the lab’s labors can be adopted.
13
Challenges of Research Portfolio Management
Keeping Focus
Get the right amounts in the right quadrants.
Don’t (let the researchers) get too comfortable.
Engage the eternal challenge of technology transfer.
IP: a two-edged sword
Offers protection from competition but can induce isolation
Senior management commitment to the portfolio in hard times
Big gains come infrequently and unpredictably.
The siren’s song of metrics
Is Corporate Computing Research Dead?
14
April 11, 2019
Tech transfer references: Clayton Christensen: The Innovator’s Dilemma, Geoffrey Moore: Crossing the Chasm
14
Let’s Get Specific!
Five examples of corporate research that had major impact on society
But what about impact on the company?
Relevance to the originating company’s business
Financial benefit to the originating company
Is Corporate Computing Research Dead?
15
April 11, 2019
With that introduction, let’s translate the abstractions into concrete cases. I’ll cover 5 examples of corporate research with major societal impact and look at how well they fit into the companies that spawned them. All of the companies involved took the long view of research and managed their portfolios with significant “upper half” components.
15
Unix (AT&T Bell Labs)
Initially developed in the early 1970s by Ken Thompson and Dennis Ritchie.
Inspired by (or a reaction to) Multics (joint MIT/GE/Bell Labs)
Commercial development ensued:
Unix V5 was licensed by AT&T commercially for $20K/instance (but there were few takers).
Non-commercial licensing to academia led to BSD (“Berkeley System Distribution”: now FreeBSD).
Commercial licensing followed (1978: DEC sold Unix/32V on VAX; many others in the 1980’s “Unix wars”).
Unix was only indirectly relevant to AT&T’s business; negligible financial impact.
AT&T sold its rights to Novell after System V release 4, around 1988
Linux – a free, open-source Unix derivative – eventually became dominant.
Redhat became a major distributor of Linux
$32B in market capitalization today, with $3.4B annual (service) revenue in 2019
Still a growth market, nearly 50 years after Unix was invented: CAGR 17% since 2015
Thompson & Ritchie won the 1983 Turing Award.
Is Corporate Computing Research Dead?
16
April 11, 2019
[Comment on the quadrant diagram on these slides: blob shows area into which the work fell, with the arrow showing trajectory over life of the system]
Thompson and Ritchie worked on Multics, which Bell Labs (and GE) worked on with MIT. Bell Labs pulled out in 1969 and Thompson/Ritchie created Unix, eliminating much of the complexity of Multics (the name being a not-so-sly dig). They wanted to use the system themselves.
Unix wars: Novell, SCO, OSF, AT&T, many others
16
System R (IBM Research)
Originated with Edgar Codd’s relational algebra ideas and 1970 paper.
Codd led a group at IBM Research.
System R was created by another group, including Jim Gray, taking a somewhat divergent technical approach and focused on SEQUEL (later renamed SQL for trademark reasons).
Prototyped 1974-77; validated 1978-79; CACM paper 1981.
IBM first shipped SQL/DS in 1981.
DB2 became a product in 1983. Still a major product 30+ years later.
DB2 had/has many competitors.
Notably Oracle (Relational Software) inspired by System R; got to market first (1979).
Edgar Codd won the 1981 Turing Award; Jim Gray the 1998 Turing Award.
Is Corporate Computing Research Dead?
17
April 11, 2019
Larry Ellison knew of Codd’s work from his 1970 paper and read about System R in the IBM Research Journal. Oracle v1, written in assembly language (!), never shipped; v2 shipped in 1979 (2 years before SQL/DS). Initial focus was DEC platforms (PDP-11, then VAX), but rewrite (v3) in C permitted wide porting.
17
Personal Computing (Xerox PARC)
PARC was created in 1970 to invent “the office of the future”.
Result: distributed personal computing
Alto, Ethernet, client-server model, file/print/email/etc. services
Xerox was slow to capitalize on the invention and was ultimately unsuccessful.
Shipped Star in 1981: business-focused, too expensive.
Apple and Microsoft (and others) adapted ideas and brought them to market.
Ubiquitous personal computing and the WIMP user interface.
Today’s LANs are essentially the same as in the 1970’s.
IP – the Internet Protocol – was modelled on PUP (PARC Universal Packets).
But Xerox did extremely well on laser printing, a component of personal computing.
Butler Lampson won the 1992 Turing Award; Chuck Thacker the 2009 Turing Award.
Is Corporate Computing Research Dead?
18
April 11, 2019
The creators of the PARC computing environments wanted to use the system themselves.
Jobs accused Microsoft of stealing from the Macintosh to make the Windows GUI. Gates (quoted by Andy Hertzfeld, who was in the room): "Well, Steve, I think there's more than one way of looking at it. I think it's more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it.“ This is accurate, as Scott MacGregor, who wrote “Windows” for the PARC D-machines, had gone to Microsoft to write the first version of Windows. Much later, he became CEO of SCO (Unix).
18
Alta Vista (DEC SRC)
Digital Equipment Corporation was a hardware company.
Founded in 1957. Pioneered minicomputers, grew rapidly, threatened IBM.
Despite rise of software, DEC remained culturally a hardware company.
Alta Vista project:
Originally intended to show off high-end DEC server hardware.
Plan: crawl 1M web pages (ambitiously large, in 1994), index with Oracle database
Indexing had a considerable research history at DEC SRC.
How to build fast, easy-to-use querying for ever-growing, dynamic data sets.
Contrast with SQL databases in usability and performance.
Alta Vista: crawler (created at sister lab WRL) plus custom indexer/query.
Initially indexed 10M pages (10x competition)
Is Corporate Computing Research Dead?
19
April 11, 2019
[Source for this slide and next: Mike Burrows]
Notes contain more detail than is appropriate for the talk, but it’s a handy place to record Mike’s recollections.
Early days of the web: browsers and sites, but no search -- just “curated” search or very limited crawling. Early search sites (1994) - Lycos, Infoseek. Paul Flaherty conceived Alta Vista as showcase for TurboLaser (AlphaServer 8400); had Oracle connection and planned to use RDBMS as index. Convinced Louis Monier to write crawler. Separately, Mike Burrows had produced NI (1990, for Usenet news), then Hector [computational lexicography with Oxford University Press] index/query facility (1992), then NI2 to do more general queries on news (1994-5).
AltaVista’s original query page became the model; adopted by Google (which added “I’m feeling lucky” and, later, the “doodle”).
19
Alta Vista (DEC SRC) [continued]
Alta Vista went live on 12/15/1995.
Experienced rapid growth in web and users, accompanied by significant growing pains.
Compaq bought DEC in 1998 for $9.8B. Alta Vista didn’t fit either company.
Compaq turned Alta Vista into a web portal (shopping, etc.), then sold it to CMGI in June, 1999.
Cost: ~20-30 man-years; TurboLasers; power, etc. => perhaps $20-40M total
Price: ~$2.4B (in stock). That’s 25% of DEC’s sale price a year earlier!
CMGI eventually refocused on search, but was crippled by dot-com crash.
If Compaq cashed out its CMGI stock, it didn’t help.
HP bought Compaq in 2001.
Is Corporate Computing Research Dead?
20
April 11, 2019
It was almost called gotcha.com, since Alta Vista was a project name and the trademark and domain were owned by others. At the last minute, “gotcha.com” was nixed by a VP who thought it was sexist, and Alta Vista went live as altavista.digital.com. Traffic grew rapidly – new httpd needed to keep up (within a few weeks) and TurboLasers were added with some regularity to try to keep up. Infoseek repeatedly attacked the site (with expensive-to-answer queries); other phenomena that looked like attacks were less certain. Whether Compaq cashed out its CMGI stock before the bubble burst is unknown (to me). Compaq was bought by HP in 2001. CMGI died a slow death (2008).
Alta Vista and DEC sale prices come from Sam Fuller, former DEC VP of Research.
20
Paxos (DEC SRC)
Leslie Lamport wrote a paper: The Part-Time Parliament
A provably correct, practical protocol for distributed system agreement
Paper was largely ignored when originally written.
Submitted for publication 1990, eventually published in 1998.
DEC made no use of Paxos in its products, though it was relevant.
Paxos is now at the heart of every modern distributed system.
Paper won the ACM SIGOPS Hall of Fame Award in 2012.
Lamport won the 2013 Turing Award, in part because of Paxos.
Is Corporate Computing Research Dead?
21
April 11, 2019
OpenVMS and RDB would have been plausible outlets. However, by the mid 1990’s DEC’s businesses were struggling. RDB was sold to Oracle in 1994. VMS customers were viewed chiefly as a legacy business; DEC’s Alphas increasingly were used to run Unix, which at the time had no real multi-processor or multi-server story.
Much more about Paxos in ACM commemorative volume on Lamport to appear in 2019.
21
Recap: Corporate research successes
All these systems had sweeping impact on society.
Opening huge new markets, spawning industries
But were they immediately relevant to the company’s business?
Yes: System R, laser printing (as part of personal computing)
No: Unix, Paxos, Alta Vista, Alto/Ethernet
Financial benefit to originating company varied greatly
Sometimes direct and substantial: laser printing (product), Alta Vista (sort of)
Sometimes indirect and substantial: System R (as DB2)
Sometimes (next to) nothing: Unix, Alto/Ethernet, Paxos
Is Corporate Computing Research Dead?
22
April 11, 2019
Each of these 5 systems had huge societal impact. The product relevance and financial benefit to the companies that spawned them was mixed.
22
Recap: Corporate research successes
It takes time! For the examples we’ve looked at:
From research concept to first demonstration: 2-4 years
From research concept to first commercialization: ~10 years
From research concept to ubiquitous use: ~20 years
On these time scales, management commitment to research is essential.
A lab takes ~5 years to build.
Most societal big wins came after 10-20 years of lab history.
Even so, no guarantee that the company will benefit!
Some societal big wins were preceded by direct benefits to company, some not.
The press doesn’t get this (and doesn’t help to build management patience).
“Paradigm shifts” are touted all the time, but actually occur rarely.
Is Corporate Computing Research Dead?
23
April 11, 2019
Sam Fuller’s 5/5/5 rule for research (from the 1990s): 5 years from idea to demonstrable prototype, 5 years to commercial product, 5 years to $1B business. Time constants have changed a bit since the 90s, but the spirit is still there (and Sam wasn’t thinking about ubiquity, which is probably another 5 years).
I would add: it takes 5 years to build a lab from scratch.
Big wins don’t come at a predictable time in the life of the lab. Bell Labs scored with the transistor (late 40’s) and Unix (70’s) but was founded in 1925. IBM Research had a similar long-term history (founded in the 40’s) and had influential results beginning in the 50’s. Xerox PARC was faster – founded 1970, Star shipped in 1981. Alta Vista and Paxos both came 10-15 years after SRC was founded. MSR was founded in 1991.
BTW: the press is focused on “the next big thing” and “paradigm shifts”. True examples come infrequently, not every year, but that doesn’t sell newspapers.
23
What About Academia?
Few highly influential systems came directly from academia.
Mach (CMU)
Picked up by NeXT, then became Apple’s iOS when Steve Jobs returned.
If Jobs had never left Apple, would it have happened?
Google (Stanford)
Gestation as PhD work/demo system
Development of advertising system occurred after start-up was formed.
As with all successful startups, the business model (not the technology) was key.
More often, academic research inspires company creation.
Akamai, RSA, VMWare, RISC, RAID, MIPS, …
Is Corporate Computing Research Dead?
24
April 11, 2019
Mach: truly intended for use and widely deployed at CMU before Jobs picked it up (at NeXT).
Google: prototype, but saw some use before Page & Brin left Stanford. Business model came later.
24
Can We Expect More Big Things From Corporate Research?
April 11, 2019
Is Corporate Computing Research Dead?
25
There aren’t many research organizations left standing: about the only two companies one can point to are Microsoft and Google, and only Microsoft has true labs. Others candidates – Facebook, VMWare, Samsung -- have business models and trajectories that don’t favor labs of the sort we’ve been looking at. In particular, such companies today seem to avoid the upper right quadrant where both risk and reward are often high.
25
From Microsoft Research?
No obvious societal “big win” yet. Too soon?
MSR was founded in an unusually small company; much early product help.
Portfolio expansion into the upper-right quadrant occurred later.
Now MSR is approaching 30 years old, so maybe…
What might be future “big wins” for society?
Custom drug design – an exploding area with early MSR involvement/expertise
Quantum computing – if it pays off, it will be a long way off, and Microsoft will be unlikely to be the “sole source”.
Privacy – if it succeeds, unlikely to produce significant revenue for Microsoft.
Universal communication – Skype translator (outgrowth of early NLP work)
Is Corporate Computing Research Dead?
26
April 11, 2019
Early days of MSR did produce some critical successes for the company, notably the Windows95/Office95 co-shipment. On the other hand, there were early opportunities that weren’t exploited that have subsequently become huge mainstream markets (e.g., MSR’s Tiger video-on-demand system in 1994).
Note that the Skype translator grew out of the natural language processing work that was one of MSR’s earliest research investments.
26
From Google?
Google deliberately avoids the research lab model.
But it has a significant research portfolio, which it must manage somehow.
What might be future “big wins” for society?
Autonomous car (Waymo – likely to be spun off)
Universal connectivity and the Internet of Things – crowded marketplace
April 11, 2019
Is Corporate Computing Research Dead?
27
Google doesn’t have a corporate research lab with the philosophy I outlined. Individual researchers are seeded around the company, not clustered. This is a deliberate choice emanating from the founders. Its ability to produce societal impacts on a scale comparable to the ones previously discussed is unproven (and, perhaps, not an objective).
Amazon is even more averse to research labs, again reflecting the founder’s view. However, they aren’t averse to disruptive innovation in business models and commercial infrastructure.
27
In Summary
Corporate computing research has had huge impacts on society, though often in unexpected ways that couldn’t be predicted when the research began.
Seminal computing systems often didn’t reward the companies that created them.
Yet big computing-based impacts on society originated more often in unfettered corporate research labs than in academic ones.
Is Corporate Computing Research Dead?
28
April 11, 2019
28
The Fundamental Quandary
Society needs the seminal systems that have come more often from corporations than from academia.
Government support for academic research has been shrinking for years.
Corporations aren’t motivated to create societal impact that doesn’t reward them financially.
“The worry I have, even at Google, is the willingness to invest in something that won’t show profits for 10-15 years is hard to come by.”
-- Vint Cerf, at South by Southwest, 3/12/2017
Corporate research labs that make substantial investments in the upper half of the quadrant diagram, though not extinct, are endangered.
So how does society continue to get the benefits that have come historically from these investments?
April 11, 2019
Is Corporate Computing Research Dead?
29
A Modest Suggestion – to You!
Commit to improve society through technological innovation.
In managing technological innovation:
Internalize the benefits of the corporate research lab approach and the importance of the “upper half”.
Seek to introduce it in the companies where you work.
Based on history, corporate research labs that emphasize the “upper half” are the mostly likely place where systems of major societal benefit will be created.
Fight to prevent them from going extinct!
April 11, 2019
Is Corporate Computing Research Dead?
30
