AWS News Blog

Amazon DevCon – Michael Tiemann

Michael Tiemann

Wrote the first native code C++ compiler, founded Cynus Solutions, now part of Red Hat.

Talking to clients about the state and strategy of open source.

At an IT technology forum last year, CEO speaker of $1B company said they would rearchitect the company around open source. IT cost and head count as % of revenue was too high. Managed to reduce these and grow revenue around open source.

Tufte, visual display of quantitative info, says you need a graph.

Executives value options, they need choices -> competition,flexibility. Oxford English Dictionary, netlib, ncbi.nlm.nih.gov, sakaiproject.org.

Cannibal principle, taking advantage ofMoore’s law. His laptop is 10x faster than the first Cray XMP. Incorporate Moore’s law into your performance and double it every 18-24 months. 30 years is 15-20
generations. Physics says that Moore’s law has 10 generations (15-20 years) remaining.

Defining enterprise architecture. Simple options encapsulating the choices to be made. Architectures for application, storage, network, identity, configuration.

Investment banking customer: Unix to Linux cost savings. $12M to $400K hardware mainteance cost. 10:1 HW cost reduction, 2:1 licensed software. HW dies, let it rot in the rack. Another customer: 3000 new servers deployed, 100TB new storage, no net new headcount cost in 2 years.

Improve TCO, imrpove collaboration expand TAM.

Quantify the value of architecture. Time to market, range of new applications, benefit of new technologies. Hardware cost savings. Amdahl’s law. Risk & risk mitigation. Cost & payoff analysis. Rational decision making.

Talking to Alan Cox, he says to look at Metcalfe’s law, Cos’x interoperability hypothesis. Degree of interoperability between competing subnets modifies the value of the network. Compatibility and cooperation increase value. Moore’s law dominates in the small, Metcalfe’s in the large. Fragmentation vs. dominance.  Inertial power of the installed base. Force multiplier. Installed base size. 1M network has 1E6 force, 20 generations or Moore’s law. 1B network has 1E9 force, 30 generations of Moore’s laws.

Network value of an any-to-any network. KDE & Gnome, decay of force with two small groups (subnets).

Standard vs. custom, good for whom? Build vs. buy.

Create subnets, lose option value since repurposing is hard or impossibe. As number of configs grow,
customization works against you.

RHEL subscriptions per quarter, now up to 150,000.

OSS improving TCO. Linux composed of lots of packages. Collaboration, development within a
commercial entity is highly silo’ed. Companies should have an OSS management and certification
program with responsibility for managing OSS. Understand licensing and legal implications, maintain
inventory, implement a good patch management system, avoid using different OSS packages to solve the same problem.

Gnome pulled in a lot of diverse packages and integrated them nicely over time.

Open source best practices, 4 freedoms.

Freedom to be different
Freedom to be the same
Freedom to modify and enhance
Freedom to share modified or original

OSS community characteristics. 300K to 1.1M developers, just 2 at OSDL. 90% of developrs are employed writing proprietary software. Most projects have 5 or less developers, rest are providing the life support to let those guys be productive. All people need to do work is permission. Developer #388 on Apache to get to 100%, top 20 guys are 80%. Most proprietary projects top off at 30-35 people. Consequence is that the marginal activity which ends up as bug fixes, downstream products, and so forth happens for OSS, not for proprietary.

opensource.mit.edu/papers/mockusapache.pdf
dwheeler.com/oss_fs_why.html
Open Source: Voices from the Open Source Revolution

OSS achieved first artlcle faster, with fewer bugs, that were fixed more rapidly.

Companies doing OSS development, hiring 1,10, etc will never get to millions. They do proprietary stuff at work, less productive, do they become stupid when they get to work? No, the model works better. RERO, modularity, cut off rough edges.

Swedish govt study,found that it really works.

Invest savings in building muscles and neurons. Collective invention and user-driven innovation.

Customers as Innovators (Harvard Business Review) Eric Von Hippel. “Unlocking the Sky“, Seth Shulman. Story of Glenn Curtiss, prolific inventor.

James Watt and steam engine, efficiency improvements, most of them happened after the patent expired, by the users.

GE plastics put a toolkit in user hands to spur innovation, give customers power to do what’s right. User driven innovation.

OSS best practices at most companies still 2-3 years out.

Crossing the Chasm book (George Gilder), sort market based on buying behavior. Rael — innovator. Thenm come Early adopters, early majority, late majority, laggards. Hit chasm between early adopters and early majority

How Red Hat does this. 1994-2003, #1 most popular distro worldwide, released twice / year. Code freeze, lots of bug fixing, 2-3 private betas, 2-3 public betas, channel release. 50% of development window closed to Red Hat, 90% to community. Update and errata model quite expensive for them.

Came up with Enterprise model to fix impedance mismatch. Get partners and customers involved, cross the chasm. 12-18 month cycle, RHL to Fedora technology model.  Profits happening, that’s proof they have crossed, since people are willing to pay.

Did Fedora project, because innovators felt left out. It is the user innovation toolkit. Development forum and proving ground for new Linux and OSS technologies. The best of what works today, no promise for the future. 100% OSS, community-driven. Roll-forward updates, not back-patch model. 7 years commitment when in RHEL.

New project, stateless Linux. Wall Street has stateless servers and stateless clients. Significantly improved reliability. Total scalability. Any app on any box at any time. No one-of configuration. Change classes of configurations,never individual machines. Made changes to Linux, FHS, LSB. Read-only root filesystem. State resides in memory, dynamic discovery, network/central server, cache. Users don’t need root. Pushing for adoption, big architectural transition. readonly-root package, rsync cache, live CD installs.

Good stuff happening on the desktop, better MSFT interop enhancements.  “Breaking Windows”, David Bank, also wrote about revolt of the corporate customer in WSJ. GNOME usability, better hardware support, stateless infrastructure. Red Hat ranked #1 in CIO survey. Desktop is ultimate stateful thing, registry is the root of all evil. Define vision that is anti-registry. Get info to right place as clients come and go, with no interruption. Win on architecture, not on better graphic design, calendaring, etc.

Q+A:

Q: Is market looking for a machine that’s like a toaster, e.g. an Apple. But people think they know what they are doing, and buy a PC.

A: Stateless desktop can help here. Perhaps GNOME desktop stateless (joking: 25 years from now).

Q: OSDL, sponsorship of developers. What does it mean that many core Linux developers work at Red Hat. How does this impact software?

A: Does Red Hat get enough credit? Maybe, maybe not. They set to connect open source software with the enterprise. Cygnus acquired by Red Hat when revenues of both were about the same. Cisco would not do acquisitions unless the acquired stuff could be ported to the GNU tools. Getting acquired gave them credibility and access, and permission to talk to the right people, foreign ministers mostly. They provide the world view, not always the “Red Hat View”. Scientific method, his distillation of what he sees out in the open world. Microsoft can never employ 1 million developers, OSS effectively does.

Fedora lets outsiders “be on the bus” as RHEL is developd. SE-Linux is a good example. The US intelligence community is all over it. Developers will finally get to do what they want in the security space.

Q: What about do no evil, and Alax Cox (not implying that he is evil)?

A: We will keep getting more people and try to make an environment to keep the ones we have. Environment, core values. Freedom, courage, commitment,accountability. Translate into do no evil.

Q: How to make work life more like OSS, and less like work.

A: He was hired to work to port a C++ code generator to some military architecture. Found that it could be done with G++ instead of using proprietary compiler. He earned the job by doing a good job on the OSS stuff. GIS is his current fascination, GRASS.

Q: Getting stuff into Fedora is still hard.

A: Agreed, we finally put resources into this in December. FUDCON, Fedora User Convention and governance model to happen soon. Before / during Linux World. Should solve many problems.

Q: Live CD?

A: Red Hat doesn’t do it, others do. Would like to see a single button in the Live core to get system to a particular configuration level.

Q: Will this OSS model apply to radically different industries?

A: Stuff he’s personally doing: Invited to Switzerland to talk to the conservation community. Create economy for indigenous people to work in their own best interests. First (bad) idea was to have the World Bank identify certain biological data, then sell options to exploit this data (silly example: tree frog excrement). Outcome was that they bought chain saws and destroyed their own forests. Now, in Brazil, you cannot legally remove indigenous species. Macau even more Draconian. This has thwarted conservation research. New thing is Science Commons and also Conservation Commons, put the biological info into this commons. Old publish or perish model protected propetary niches of scientists and publishers.Showed them the OSS model and convinced people to share more information, unanimously declared that they would launch Conservation Commons. Open source information facility. Similar stuff in pharmaceuticals. OSS saved human genome and kept it out in the open, vs. what Celera tried to do. Welcome trust bought a bunch of Linux boxes and out-sequenced Celera, put info into public domain.