How Open Source Works Michael Tiemann
SIIA Webcast
Agenda
• What is Open Source? • How is Open Source Developed? • How Does Open Source Development Scale? • How Does Open Source Fit into the Enterprise? • Why You Should Care
SIIA Webcast
Open Source Is...
• Any software that satisfies the Open Source Definition (see http://opensource.org/docs/definition.php)
• Licenses approved by the OSI can be found here: http://opensource.org/licenses • The General Public License (GPL) is an open source license which covers over 70% of all Free and Open Source Software (FOSS). • TCP/IP, the World Wide Web, the Apache web server, Mozilla, Linux, OpenOffice, Perl, PHP, and JBOSS are all well-known open source projects
SIIA Webcast
Open Source Developers are...
• Everywhere • Community size is estimated to be over 200,000 programmers • The Linux Kernel has over 200 “maintainers” representing over 300 organizations
• More than 10,000 programmers have contributed to Linux • Open Source is the ultimate “opt-in” model:
“When a private individual mediates an undertaking, however directly connected it may be with the welfare of society, he never thinks of soliciting the cooperation of the Government, but he publishes his plan, offers to execute it himself, courts the assistance of other individuals, and struggles manfully against all obstacles. Undoubtedly he is often less successful that the State might have been in his position; but in the end the sum of these private undertakings far exceeds all that the Government could have done.” -- Alexis de Tocqueville, Democracy in America
SIIA Webcast
1999: “Linux Doesn't Scale”
• Intel-based hardware decidedly “low-end” • Effectively limited to 2-way scalability (BKL) • No large-memory support • No two enterprise applications run on any one version of Linux • Microsoft “benchmark” shows IIS on Windows is 100x faster than Apache on Linux (nb. SPECweb99 performance, not how quickly security compromises are found/exploited)
SIIA Webcast
2000: Linux does scale
• Zero-copy TCP/IP network stack • Breaking of the BKL • Intel-based hardware beats proprietary hardware performance for edge, workgroup computing
• Red Hat and Dell set 5 SPECweb99 benchmarks: • 1 cpu • 2 cpu • 4 cpu • 8 cpu • Unlimited cpu
SIIA Webcast
2001: Linux better scale!
• IBM commits $1B and hundreds of developers to Linux • SMP scalability • Threading • Volume management • Intel-based hardware beats proprietary hardware for 90% of all enterprise computing requirements
• ReiserFS and ext3 journaling filesystems • Enterprises begin to look seriously at Linux • Analysts ask “how far can Linux scale?”
SIIA Webcast
The Linux Kernel Scheduler
• Only bit of Linux to remain from original kernel • “GOOD CODE!” • July 2001: IBM proposed multi-queue scheduling approach showing 40% scheduling improvement on some benchmarks
• October 2001: Linus just says “NO” • January 2002: Ingo Molnar is “bored”, releases O(1) patch • January 2002: Linus accepts patch
SIIA Webcast
Studies in Scalability (1)
1000 Context Switches/Second
6000 5500 5000 4500 4000 3500 3000 2500 2000 1500 1000 500 0 2 cpu 8 cpu
Orig. O(1)
Benefit to real-world trading application: 20% net speedup!
SIIA Webcast
Studies in Scalability (2)
• Security: LSM • Database: AIO – 138K+ TPC/C (Oracle 9i on 4-way Proliant) • Threading: NPTL vs. NGPT • 100K threads in 1 second (2-way) • 1M threads in 20 seconds (24-way)
SIIA Webcast
Scalability is a feature to preserve
• One step back for two steps forward: no good • Patches which limit options: no good • Democracy among developers has avoided tyrrany of the proprietary model
• Proof: IBM, Oracle, Intel, HP, Sun, others all contributing to Linux, open source without slowing open source down • A bit like the Internet: the more there is, the more valuable it becomes
• A bit like Google: scalability through massive inclusivity
SIIA Webcast
The Proprietary Software Crisis
● ●
Faulty software costs businesses $78B/year Traditional software model actually encourages buggy products
●
Lack of competition has led to vendor intransigence
Proprietary software often results in bad architecture Bad architecture locks you in
SIIA Webcast
$78 Billion vs. S&P 500 Earnings
Quarterly Earnings (derived from EPS)
140 130 120 110 100 90 80 70 60 50 40 30 20 10 0 1999 1999 1999 1999 2000 2000 2000 2000 2001 2001 2001 2001 2002 2002 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2
$78B As Reported Reported+Good SW Operating
Opportunity
Cost
SIIA Webcast
Quantifying the Value of Architecture
● ●
Time to market (10% vs. 20% vs. 30%) Range of new applications (90%) Effective benefit of new technologies (90%)
●
●
Risk and Risk Mitigation (10%)
Savings $M
125 100 75 50 25 0 Year 1 Year 2 Year 3 Year 4 Year 5
A: 10% replacement/year
B: 20% replacement/year
C: 30% replacement/year
SIIA Webcast
Quantifying the Value of Architecture
● ●
Time to market (30%) Range of new applications (50% vs. 70% vs. 90%) Effective benefit of new technologies (90%)
●
●
Risk and Risk Mitigation (10%)
Savings $M
125 100 75 50 25 0 Year 1 Year 2 Year 3 Year 4 Year 5
A: 50% max replaceable
B: 70% max replaceable
C: 90% max replaceable
SIIA Webcast
Quantifying the Value of Architecture
● ●
Time to market (30%) Range of new applications (90%) Effective benefit of new technologies (60% vs. 75% vs. 90%)
●
●
Risk and Risk Mitigation (10%)
Savings $M
125 100 75 50 25 0 Year 1 Year 2 Year 3 Year 4 Year 5
A: 60% optimal
B: 75% optimal
C: 90% optimal
SIIA Webcast
Quantifying the Value of Architecture
● ●
Time to market (30%) Range of new applications (90%) Effective benefit of new technologies (90%)
●
●
Risk and Risk Mitigation (30% vs. 20% vs. 10%)
Savings $M
125 100 75 50 25 0 Year 1 Year 2 Year 3 Year 4 Year 5
A: 30% risk
B: 20% risk
C: 10% risk
SIIA Webcast
Quantifying the Value of Architecture
● ●
Time to market Range of new applications Effective benefit of new technologies
●
●
Risk and Risk Mitigation
Cumulative Savings ($40M annual spend, 3 Scenarios) 125 100 75 50 25 0 Year 1 Year 2 Year 3 Year 4 Year 5
A: 10% per year to 50% max 60% optimal, 30% risk
B: 20% per year to 70% max 75% optimal, 20% risk C: 30% per year to 90% max 90% optimal, 10% risk
SIIA Webcast
18 Months, 100 Million Validations
“[Jeff Birnbaum]'s busy replacing 4,000 highpowered servers running traditional software with much cheaper machines running Linux. Projected five-year savings: up to $100 million.” Business Week / March 3, 2003
% of Intel CPUs
100 80
60
40
20
0 2001 2002 2003 2004 2005 2006 2007
Source: MorganStanley
SIIA Webcast
The New Enterprise Computing Mandate
●
Enterprises insist on open standards
●
They insist on retaining their power of choice They insist on flexible solutions "We came to believe that a power greater than ourselves could restore our IT department to sanity. Linux is that power."
Scott Berinato, CIO Magazine January 2002
●
SIIA Webcast
Why Open Source Matters
• Historic proprietary platform model has run its course • Bad architecture/proprietary lock-in has frozen out innovators • Open Source platforms with strong support for open standards provide a level, liberated, playing field
• Open Source expands the market by giving IT buyers more choices, more compatability, and more return on their investments
SIIA Webcast