‘‘Tw o years in the trenches’’ Evolution of a free software project

GregLehey President, AUUG Inc. [email protected] Member,FreeBSD Core team [email protected]

ABSTRACT Free software projects have been around for overten years now, and they’ve become an important part of the UNIX community.With the example of the FreeBSD project, this paper illustrates some of the changes that have occurred in that time and attempts to guess what the future may bring. The three ages of UNIX The evolution of free UNIX

UNIX® is nowathird of a century old. In this Free software has been around for ever. Until the time, the face of computing has changed dramati- end of the 60s, coincidentally the time when cally.It’sinteresting to consider the evolution of UNIX evolved, software was almost always free. UNIX in three phases of roughly 11 years each: And whynot? Most system software came with the machines and would only run on them, so • From 1969 until about 1980, UNIX was there was little incentive tocharge separately for mainly a research project, little known outside the software. Applications software was tuned to AT&T except at some universities. specific applications, so piracywas hardly worth • From 1980 until about 1991, UNIX developed worrying about. into a commercial operating system with re- By the end of the 1960s, that had changed. IBM leases likeUNIX System V (remember that?), wasfaced with competitors who built hardware XENIX, SunOS, Ultrix and friends. which was strikingly compatible with the Sys- • In the early 90s, efforts at the University of tem/360, and IBM’ssoftware ran on it. It started California in Berkeleytoproduce free UNIX to makecommercial sense for IBM to unbundle came to fruition. At pretty much exactly the its software. Over the course of the following ten same time, the Linux project started. UNIX years, more and more compatible hardware be- wasonthe way to becoming free. came available, and by 1980 just about all soft- ware cost money. Vendors had a perception that This paper looks at the last third of this history in access to the source code would give their com- more detail. petitors an advantage, so theyrestricted availabili- ty of their software to object form only.The re- sultant inflexibility was one of the reasons which led to the formation of the Free Software Founda- tion in the mid-80s. To wards the end of the 70s, the price of ‘‘real’’ computers dropped to the point where individuals • The release of the Intel 80386 processor and could own them. People started sharing programs systems based on it, gav e the average hacker for their CP/M based machines, and later for their an affordable machine with virtual memory, Microsoft-based replacements. Most software for capable of running modern UNIX without small machines continued to cost money, howev- significant compromises. er,and the machines themselves were too small to • Improvedaccess to the Internet, especially for run UNIX effectively. students, made cooperation on software pro- This changed round about 1990 with the availabil- jects more practicable. ity of machines based on the Intel 80386 proces- In those days, for most people the attraction was sor.Atthe same time, individual access to the In- the challenge of running UNIX on one’sown ternet became easier,especially for students. computer,not anycommercial intent. Indeed, These were the background for a number of free Bill Jolitz staunchly refused to commercialize his software projects which were initially completely 386BSD operating system, and in August 1991, independent of the Free Software Foundation and Linus Torvalds wrote in the original announce- its values. ment of Linux ‘‘just a hobby,won’tbebig and The early 90s professional likegnu’’. Idon’tbelieve that religious belief in ‘‘free soft- Computer hackers1 have been around as long as ware’’was a big thing at the time. The GNU pro- computers. Theyhav e probably dreamed of hav- ject had been around for a while, and both 386 ing their own computer that long, too, but it really BSD and Linux took advantage of the software, only became evenremotely possible in the butinitially there was little synergy between the mid-1970s. People started working on UNIX-like GNU project and the free OS projects. systems early: During this period, relatively fewpeople con- • In the late 1970s, a companycalled Electro- tributed to the projects. Forexample, the release labs started working on a UNIX-likesystem notes of FreeBSD 2.0, released in January 1995, called OS-2, intended to run on the Z-80 pro- mention a total of 55 contributors, including a 15 cessor.Itnev erpassed the beta stage running member core team, who really were the project. under CP/M. Most of the other contributors had only loose • In the early 1980s, Mark Williams Company links with the FreeBSD project. ported their Coherent system, a clone of the UNIX Seventh edition, to the IBM PC. With- The compromise OS out memory protection, and with the memory Along with the free operating system projects limitations of the original success, it was not a there was also a ‘‘reasonable’’operating system great success. project, BSD/386 from BerkeleySoftware Design • Andy Tanenbaum’sMinix operating system Inc., or BSDI (later BSDi). BSD/386 was derived ran on a number of processors, including the from the same code base as 386BSD, and was in Intel 8086 family. fact quite closely related. The story of that rela- tionship is interesting enough to be the topic for None of these projects became very large. The another paper.The difference was that BSD/386 hardware wasn’tuptoit, and none of them made cost money. Including complete source code, it the source code freely available, though the re- cost $1,000 US. That may seem to be a lot of strictions on Minix were relatively minor.Two money, but in those days source licenses for Sys- things changed that situation in the late 1980s and tem V cost at least $50,000, and a BSD/386 early 1990s: source license was was cheaper than a binary li- cense for the System V implementations for the Intel platform. In addition, it was complete—no dozens of additional packages to add, each with their own activation key.Evenmore surprisingly, 1. Ahacker is, of course, somebody who can’ttell the difference between work and play with computers. See the BSD/386 was more reliable than System V. NewHacker’sDictionary (http://www.tuxedo.org/˜esr/jargon/html/entry/hacker.html ) for more details. The mid-90s In the early days, nobody had very great expecta- Social changes tions for the free operating systems. Theydidn’t expect them to be as reliable as UNIX® System Clearly the social structure of the FreeBSD pro- V. Theydidn’tevenexpect them to be as reliable ject has changed greatly in its nature in the last as Microsoft’sofferings, though theydid expect ten years. By mid-2000, the strain was beginning them to be easier to use. to show: In a fewyears, something happened that nobody • The core team was no longer the group of the had expected. Well, very fewpeople: the operat- most active comitters. Their function had ef- ing systems did become as reliable as Microsoft’s fectively become more administrative,but offerings, quite quickly in fact. One of the first theyhadn’trecognized the fact yet. results of this recognition came from the massive • The architectural direction of the project had increase in Internet usage, which created a re- become a little vague. The position of chief quirement for low-cost web servers. This wasthe architect, previously held by David Green- reason for the foundation of BSDI, who branded man, had been vacant for some time. In the their operating system ‘‘Internet Server’’, later early stages of the project, most of the work ‘‘Internet Super Server’’. Initially theywere very in the project had been to makeFreeBSD a successful. As time went on, though, enterprising stable UNIX-likeoperating system, but that startups realized that theycould save the cost of goal had nowbeen achieved. There wasmore the software by using free operating systems in- work to do, but the only clearly defined goal stead of BSD/386. Thus in 1995, when Yahoo! wastoimprove the SMP performance—the started up, it was based on FreeBSD. SMPng project, which at the time had just Linux was slower to reach reliability than the started, and which is still continuing. BSD systems: it had to be written from scratch, • Arelated problem was the attractiveness to whereas the free BSD systems were based on a end users. Likeother free operating systems, code base at least ten years old, including the FreeBSD has always been a developer-driven most mature of TCP/IP stacks. project, but the main source of project fund- By the late 1990s, though, Linux had effectively ing came from selling CD-ROMs. The sales caught up with the BSDs, and depending on of the CD-ROMs were obviously dependent which bigot you ask, either overtook the BSDs or on the perception of the purchaser,but no- neverquite made it. body in the project was overly concerned with this issue. In the mid-1990s, another thing happened: the general public became aware of the concept of • Some developers exploited the ineffectiveness free operating systems. Theywere still consid- of the core team by doing whateverthey ered very much the realm of geeks, but theywere wanted. This notably included making becoming known. changes to the system to match their viewof what was needed, possibly breaking parts of During this time, more and more people became the system (those parts which theydidn’t involved with the projects. use), and leaving it to others to clean up. The late 90s • This in turn, along with an observed inability on the part of the core team to solvethe prob- By the end of the 1990s, free software was well lem, caused a serious decline in the morale of enough known in the IT industry for some people the project, and a number of people left the to form companies to market it. The result was a project as a result. further increase in the size of the projects. At the time of writing, the FreeBSD project has 318 • One of the most noticeable rogue developers committers, in other words developers with write wasamember of the core team. As a result, access to the source tree, probably ten times the the lack of activity of the core team was per- number as of January 1995. ceivedascronyism. • The core team had adopted a policyofnot • Don’tchange anything, core is fine the way it communicating the reasons behind its deci- is. Received7votes sions, partly to hide the fact that manymem- • Disband core entirely and let committers cre- bers were inactive,and partly to avoid spark- ate a newstructure in its place. Received7 ing conflicts. Not surprisingly,this gav e them votes areputation for secrecy. Clearly the majority was for a democratically In November 1999, Nate Williams asked in a mail elected core team. More discussions ensued. message: Some people were concerned that politics would Finally,what is the purpose of core? I used to takeoverfrom reason, and the people who would know,but I'm not sureanymore. What determines get elected would be those who could drum up if someone should become a coremember? Is enough followers, not those who could do the best thereany way to lose your coremember status, in jobs. A team of volunteers, consisting of the same manner that you can lose the ability to Jonathan Lemon, Warner Losh and Wes Peters, be considered a maintainer? Do you have to quit got together and thrashed out the existing voting in order to not become a coremember? (So far models and came up with the following proposal, that'sthe case). on which we also voted: My *biggest* fear is that we will lose active de- • Active committers have made a commit to the veloperssimply because we just plod along hop- tree in the last 12 months. ing that everything will work out, and hope that • Core consists of 9 elected active committers. someone will pickupthe torch fromtime to time and takeusinsome sort of good direction. • Core elections are held every 2 years, first time September 2000. Lah-de-dah, lah-de-dah. Once upon a time,core memberswerefolks were*really* excited and • Core members and committers may be ejected highly motivated to work on this thing,and would by a 2/3 vote of core. spend nights/weekends and all sorts of time on • If the size of core falls below7,anearly elec- this. But, coreisnow older,and our real lives get tion is held. in the way now. • Apetition of 1/3 of active committers can Acouple of people suggested various ways to re- trigger an early election. form or change the core team, including the possi- bility of disbanding core altogether and becoming • Elections will be run as follows: an anarchy, orvoting for the core team. Anum- Core appoints and announces someone to ber of people came up with remarkably compli- run the election. cated voting models. Finally,Jordan Hubbard 1week to tally active committers wishing to came up with a suggestion and asked the develop- run for core. ers to vote on a number of alternatives. Out of 94 4weeks for the actual vote votes cast, the most popular were: 1week to tally and post the results. Each active committer may vote once in • The idea of core is fine, its membership sim- support of up to nine nominees. ply needs a shake-up and some mechanism Newcore team becomes effective 1 week added for voting in newblood. This alterna- after the results are posted. tive received58votes. Voting ties decided by unambiguously elect- • The idea of core is fine, but some of members ed newcore members. simply need to leave.This received12votes, • These rules can be changed by a 2/3 majority most of which identified a single specific of committers if at least 50% member. of active committers cast their vote. • Core needs to be broken up into an over- These ‘‘bylaws’’passed by 117 yes votes to 5 no sight/human resources group, leaving archi- votes, thus also disproving the concern that com- tectural decisions to developers. This alterna- mitters wouldn’tbeinterested enough to vote for tive received9votes. the core team. Acouple of these provisions look a little unusual: • The rationale behind the surprisingly long • Warner Losh, newly elected. Network hacker. election period was that, since this is a volun- American. teer project, manypeople might miss a shorter • Doug Rabson, member of the old core team. election period, especially Europeans on mul- Kernel hacker,responsible for the port of ti-week leave. FreeBSD to the Alpha platform. British. • We spent a lot of time discussing howtovote. • MikeSmith, newly elected. Low-levelkernel We were concerned that if each voter had only hacker.Australian (Adelaide). one vote, the majority would vote for the same twoorthree candidates, effectively leav- • Robert Watson, newly elected. Network ing the remainder to chance. Even worse, it hacker,FreeBSD security officer.British. could lead to less than 9 candidates being vot- • Peter Wemm, member of the old core team. ed for at all. Initially,wealso discussed a Universal Kernel Hacker.Australian (Perth). ‘‘veto’’vote: ‘‘Don’tlet onto core’’. You’ll note from Jordan’spoll that the In summary,the newteam included fivemembers second most popular model was ‘‘expel from core’’. Neither of these sugges- old core team were not re-elected. The composi- tions were accepted. tion of the team changed in other ways: fivemem- bers of the old core team had a non-English native The election language, but only one member of the newteam did. Sevenmembers of the old core team lived Nominations for candidacywere accepted be- outside the USA, only twoofthe newteam did. tween 5 September 2000 and 12 September 2000, Three members of the newcore team were Aus- after which the election started immediately.It tralians, including myself, compared to twobe- finished on 10 October 2000. The results were fore. announced on 12 October 2000, just in time for the beginning of BSDCon 2000 One thing that all members had in common was (http://www.bsdcon.com). that theywere software developers, not managers. This is not surprising, giventhe mode of election. Atotal of 17 candidates registered, surprisingly Ihad had some management experience years be- close to the size of the previous core team. Only fore, but I believe that I was the only one, and my 8ofthe previous core team stood for election. experience wasn’tmuch to write home about. Campaigning was almost non-existent. The morning after Into the trenches The members of the newteam, later to be dubbed The newcore team took office at a panel discus- core.2,were: sion during BSDCon 2000. We had a completely newconcept ahead of us, and we certainly • Satoshi Asami, member of the old core team. weren’tsure howtofulfil our objectives. Worse, Guardian of the Ports Collection Japanese. we didn’tevenknowwhat our objectiveswere. • David Greenman, one of the founders of the An association likeAUUG has a constitution. FreeBSD project, and member of the old core The best we had in the FreeBSD project were the team. Kernel hacker and former principal ar- ‘‘bylaws’’, originally intended to define the chitect of the FreeBSD project. American. modality of the elections. • Jordan Hubbard, one of the founders of the The first meeting FreeBSD project, and member of the old core team. Release engineer and former president The second FreeBSD core team had a meeting at of the FreeBSD project, a position which he the end of the BSDCon in Monterey. Itwas, in had dropped some time before. American. fact, a significant event: in the course of the histo- ry of the FreeBSD project, it was the only meet- • GregLehey, newly elected. Kernel hacker, ing of the entire core team. Previously,anattempt author of the Vinum Volume manager.Aus- to pay to bring the core team together (at the tralian (Adelaide). FreeBSDCon in Berkeley, the year before) had failed thanks to the efforts of the US Immigration Department: AndreyChernov, who livesinRus- • Rogue developers. Wecouldn’tagree on how sia, was deemed unsuitable for entry to the USA. to handle this one. One of the issues that was made very clear was that the core team did In this meeting, we tried to define what the pur- not have a ‘‘big stick’’. About the only thing pose of the FreeBSD core team was. Wediscov- that it could do would be to expel a member ered a surprising number of differences of opin- from the project. ion, but we finally decided: • Project morale. This included behaviour of • The FreeBSD core team does not define the developers towards each other.Again, core architectural direction of the project. did not come up with a good solution for this • There will be no officers on the core team. problem, though theoretically expulsion from Jordan Hubbard had suggested to takethe role the project would have been a solution. of spokesman, but the consensus was that people already sawhim as exercising too Acceptance of core.2 much control on the team, and we suspected this would send the wrong message. Howdid we go? Parts of it were excellent. • The FreeBSD project is a volunteer organiza- The biggest problem we found was that core tion, so the core team does not have a man- members were still unresponsive.Applications date to tell anybody to do anything. for commit bits, our main activity,took up to sev- eral months to process instead of the one-week That’sconceding a lot. So what was left? timeout we had set ourselves. One of the prob- • The core team decides who can join the pro- lems was the amorphous structure of core: no- ject as a ‘‘committer’’, somebody with com- body had a particular hat, so there was nobody mit access to the CVS tree. On request, designated to actually convey the message back to backed by a mentor,who must already be a the applicants. In the course of time, that got bet- committer,the core team decides whether to ter.Very fewapplications were rejected. admit him to the project (to ‘‘give him a com- Publishing the monthly core reports became very mit bit’’). Core voted on each application. A slow. Although the reports for the last months of single ‘‘no’’was sufficient to veto an applica- 2000 were relatively timely,the January 2001 re- tion, and voting terminated after a week. port was released on 29 June 2001. We beg anto • In case of extreme misbehaviour,the core recognize that we needed help, and solicited ap- team can expel a committer from the project. plications for the position of core secretary,anon- voting position more akin to AUUG’sbusiness • In case of dispute between twocommitters, manager than to the secretary.Initially we let the core team mediates. several applicants work on the backed-up reports. Alittle later we added the concept of a monthly The February 2001 report appeared in November core team report to address the accusations of se- 2001, and gradually we caught up with the back- crecymade against the previous core team. log. Wesigned up WilkoBulte as core secretary, aposition he still holds, and by May 2002 we had Comparing this list with the problems facing the cleared up the backlog. project, a number of issues remained unanswered: • We still had no architectural direction. The Rogue developers core team’sintention here was that a consen- Not surprisingly,problems with rogue developers sus should be formed on the FreeBSD- did not abate. Each occurrence caused a lot of an- arch mailing list. If no consensus could be gry discussion with core, which was very wearing formed, core would mediate. on a number of the members. Surprisingly,italso • Attractiveness to end users. The majority of became apparent that manycore members saw the members of the core team, being develop- each occurrence as a separate issue, and personal ers themselves, were not very interested in likes and dislikes were very evident. There was this aspect. strong resistance to anygeneral policy. In February 2002, a developer announced his in- tention to commit some significant changes to the SMP code. At the time, the most active SMP de- pended for 5 days. Any member of corecan veloper,John Baldwin, was moving house from implement the suspension without the need one coast of the US to the other,and was thus of- for a formal vote within core. The suspen- fline. Others who were involved pointed out that sion will be published on -developers. these changes were in conflict with changes that 5. Corereserves the right to impose harsher John was currently testing and asked the develop- penalties for repeat offenders. Harsher er to hold off. The developer committed the penalties include longer suspension terms changes anyway. and the permanent removal of commit priv- The handling of this particular issue became a test ilegesand FreeBSD.orgaccounts. Imple- of core’sauthority.For the first time, core decid- menting harsher penalties aresubject to a ed to revoke the developer’scommit privileges if formal 2/3 majority vote in core. The out- he did not back out the commits. He did so in the come of core’sdecision will be published nick of time (without knowing about the impend- on -developers. ing suspension), and asked core to resolvethe is- In all cases whereanindividual FreeBSD officer sue. The resolution was hard, and it looked more takes a personal action he or she will be answer- liketactics rather than strategy.After a month of able to core. All cases can be taken for appeal to discussion involving hundreds of mail messages, the coreteam. The outcome of suchanappeal will core appointed John Baldwin to the position of be published on -developersand in the core technical lead for the SMP project, with the pow- monthly report. er to approve orreject changes. These rules looked rather rigid, but we couldn’t Based on this relative success, core deliberated come to an agreement to moderate them, so that’s and came up with some rules about developer be- the way theyremain. haviour: 1. Any committer who commits to the stable The big stick branchduring a code freeze will have his or In June, core receivedaformal complaint about her commit bit suspended for 2 days. Any the same committer who had caused us so much member of coreorthe re@ team can imple- grief in February.Hehad committed code in an ment the suspension without the need for a area on which another developer was working, formal vote within coreorre@ respectively. without discussing the matter with him. This had The suspension will be published on -devel- annoyed the other developer to the point that he opers. relinquished the maintainership of this part of the 2. Any committer who commits to the security tree. branchwithout approval from the security- We discussed the matter and attempted to decide officer team will have his or her commit bit whether this behaviour was in conflict with the suspended for 2 days. Any member of core rules we had just published, specifically rule 4. A or the security-officer@ team can imple- majority decided that it was, but there were exten- ment the suspension without the need for a uating circumstances. According to the rules, formal vote within coreorsecurity-offi- though, we still had to impose the full fiveday cer@ respectively.The suspension will be penalty. published on -developers. The developers reacted in different ways, mainly 3. When committersengage inacommit war, unfavourably.Inthe meantime, as described the both parties will have their commit bits sus- next section, we were in an election campaign. pended for 5 days. Any member of corecan Some suspected that this punishment was politi- do this without the need for a formal vote. cally motivated, since the developer in question The suspension will be published on -devel- wasalso an election candidate. It’snot clear, opers. however, whether this punishment improvedor 4. Any committer observed to act or speak in lessened his chances, but in anycase he wasn’t away that is in conflict with the normal elected. The core team decided on a reprieve af- rules of interpersonal politeness, or in con- ter twodays, and the matter died down relatively flict with the best interests of the FreeBSD quickly. Project will have his or her commit bit sus- Afew weeks later,two other highly respected have a lot of preconceived notions of what coreis committers engaged in a commit war: one com- to makethat happen. mitted something that the other didn’tlike, the Finally,italso bearsnoting that while being part other backed it out, the first recommitted it, and of the FreeBSD project is many things, it should so on—a clear violation of rule 3. As always, always be "fun" to at least some degree for its there were extenuating circumstances. After participants or there’sreally not muchpoint in some deliberation, core decided to suspend the being involved. Being in core, whereone gets to commit bits for 24 hours. Again, this caused a deal almost solely with conflict resolution and bu- commotion in the mailing lists, but it died down reaucracy,isnot fun in any sense of the wordand more quickly. while being in coreconstitutes the bulk of my in- Is this working? It seems to. Core needs to un- volvement, without any cool development work derstand howtodose the punishment, but the real (whichIalso haven’thad time for) to counter-bal- issue here is not the temporary loss of commit ance it, it simply leaves me with less and less en- privileges, it’sthe open recognition of inappropri- thusiasm for FreeBSD. ate behaviour.It’sstill too soon to be certain, but Yes, it certainly wasn’tfun. Slashdot picked it up maybe people are being more considerate as a re- with glee, of course, and the usual ‘‘FreeBSD is sult. dying’’trolls came out again. That’snot the The collapse of core.2 point, of course. This has nothing to do with FreeBSD as an operating system, these are simply Round May 2001, Satoshi Asami became sick some interesting project dynamics. and disappeared from the scene for some time. As if that wasn’tinteresting enough, fivedays lat- Even after his return, he did not participate in core er MikeSmith also resigned from core. In his discussions, and after several months, we finally message, he wrote: decided that he was de facto no longer a member of the core team. According to the ‘‘bylaws’’, we Fr eeBSD used to be fun. It used to be about do- carried on with only eight members. ing things the right way.Itused to be something that you could sink your teeth into when the mun- After the SMP commit war described above,peo- dane chores of programming for a living got you ple were feeling tired. Everything seemed to take down. It was something cool and exciting; a way more effort than necessary.On29April 2002, to spend your sparetime on an endeavour you Jordan Hubbard dropped a bombshell: he re- loved that was at the same time wholesome and signed from core. In his resignation message, he worthwhile. stated: It’snot anymore. It’sabout bylaws and commit- ... being in coreishonestly not what it once was. tees and reports and milestones, telling others Foranold-timer likemyself,who was used to a what to do and doing what you’re told. It’sabout coreteam that was far morecohesive and gener- who can rant the longest or shout the loudest or ally on the same page, it’ssimply a painful expe- mislead the most people into a bloc in order to le- rience a lot of the time.Perhaps this is due to gitimise doing what theythink is best. Individuals overly rose-colored recollections of the old core notwithstanding,the project as a whole has lost on my part, and I do certainly recall us having trackofwhereit’sgoing,and has instead become morethan our shareofdisagreement and ineffi- obsessed with process and mechanics. ciency in the past, but on the balance corestill ... feels too muchlikethe pre-WWII Polish Parliment Fr omatechnical perspective,the project faces a sometimes, wherewe’re fully capable of arguing set of challenges that significantly outstrip our some issue right up to the point wheretanks are ability to deliver.Some of the resources that we rolling through the front door and rendering the need to address these challenges aretied up in the whole debate somewhat moot. I’m also not blam- fruitless metadiscussions that have ragedsince we ing this on the democratic model we’ve adopted, made the mistakeofelecting officers. Othershave astance whichwould be hypocritical at best since left in disgust, or been driven out by the cultureof I’m one of the folks who argued strongly in favor abuse and distraction that has grown up since of it, but I guess it’sgoing to takeafew moreiter- then. Moremay well remain available to recruit- ations beforeweget it right. It will also probably ment, but while the project is busy infighting our be a lot easier for truly newpeople who don’t chances for successful outreachare sorely dimin- • Robert Watson, member of core.2. Network ished. hacker.British. Does this look familiar? It should do to anybody • Peter Wemm, only member of the original in commercial software development projects. core team left. Universal Kernel Hacker. Coincidentally,both Mikeand Jordan work for Australian (Perth). Apple. Theymust be involved in project planning This time around, all those members of the sec- there; it’stobeexpected that it works better at ond core team (four) who stood for election were Apple. re-elected. core.3 In manyways, this core team composition is the best we have had. In particular,wehav e better After MikeSmith’sresignation, core only had six technical representation (John Baldwin) and the members left. According to the ‘‘bylaws’’, this representation of the Release Engineer,Murray meant that elections had to be held. That, too, Stokely. caused long discussions. When, howquickly, The first elected FreeBSD core team got offtoan should we change the ‘‘bylaws’’first? Wedid a erratic start. As the resignation letters of Jordan strawpoll which showed that the committers did Hubbard and MikeSmith show, this was at least not want to wait, and theydidn’twant to change partially because of unrealistic expectations of the the bylaws. An election schedule was published tasks involved. The newcore team looks likeit in accordance with the ‘‘bylaws’’, and then people might finally be pointing in the right direction. decided there wasn’tenough time for people to declare their candidacy. Accordingly,the period for nominations was extended, and voting ran Conclusions from 29 May to 25 June. The idea of independent free software projects is Despite the perceivedtiredness, a record number still very new. Things are changing rapidly,and of nominations were received. In contrast with it’sdifficult to guess what will happen in the fu- the first core elections, there was a significant ture. A number of things have become evident, amount of politicking, with some candidates pub- though: lishing lists of their preferred partners in core. • It’spossible to run a small software project The results were announced, not as planned on 1 without significant administrative overhead. July,but immediately after the elections closed. It’simpossible to run a large software project The third core team consists of the following without significant administrative overhead. members: • Agood kernel hacker is not automatically a • John Baldwin, newly elected. FreeBSD SMP good manager. technical lead. American. • The problems that large projects face do not • Jun Kuriyama, newly elected. Japanese. differ significantly between volunteer and commercial projects. • GregLehey, member of core.2. Kernel hack- er,author of the Vinum Volume manager. • Working on a big software project has almost Australian (Adelaide). neverbeen ‘‘fun’’. That wasone of the rea- sons that the free operating system projects • Warner Losh, member of core.2. Network started in the first place: here was a chance to hacker.American. work on software free from the constraints of • Mark Murray,newly elected. Security hacker. project management. This aspect, sadly,isa Zimbabwean. thing of the past. • WesPeters, newly elected. Network hacker. In summary,Isuspect that the FreeBSD project, American. and other similar projects, will gradually become more formalized, more likecommercial operating • Murray Stokely,newly elected. FreeBSD Re- systems. So what will distinguish them? Ifear lease Engineer.American. that the distinctions will become less and less as time goes on.