The march to open source: LinuxWorld San Francisco 2005 wrapup
by Andy Oram
Related link: http://linuxworldexpo.com/live/12/events/12SFO05A
A different sort of evidence of free software's forward march came
today at LinuxWorld. I'll report on that here, along with some other
interest events of the past two days:
My earlier blogs from this show are:
The march to open source
Two sales people (the kind of staff most evident at LinuxWorld) from
two different companies approached me within a 24-hour period to say
they want to talk to me about making part or all of their product open
source. I was flattered to be consulted, being neither a system
designer nor a marketing expert. But before I directed them to others
I thought would have more direct experience, I had some interesting
conversations that laid out some of the issues.
The determination of a company to open-source its product is at least
as significant as the decision of a customer to use open-source
software. When a company entrusts its core value to the community, and
declares that it can make a living by doing so, the movement has a
major gain. And it's much harder to open a proprietary product after
building a company around a proprietary model than to start out open
source. These two companies--Radiant
Qlusters--are brave folks.
Morality versus practicality
I don't for a moment believe that, when leaders of the free software
community met at a
summit organized by Tim
O'Reilly in 1998
and chose to promote their work as the "open source" movement, they
were abandoning ethical or moral beliefs about the value of free
software. Nor do I think anyone who supports free software believes it
lacks practical application. But somehow an unwholesome view has
arisen that there are two different positions on free software, one
saying, "We wouldn't care if free software had no practical value,
we'd insist on it anyway" and the other saying, "Go open source and
we'll make you rich."
But there are certainly times when one must decide what arguments to
stress when promoting free software. I chose the moral one in my 2002
Why Human Rights Requires Free Software. In talking to Radiant
Data and Qlusters, I chose a more practical approach.
These companies have to deal with investors, partners, VARs, and
customers with critical needs. Few of these will find it persuasive if
a company says they made their code base free because "We like open
source" or "We want people to cheer us." The people on the other side
of the bargaining table want to see tangible benefits. Placed suddenly
on the front lines, therefore--asked point-blank to help a company
justify going free--I found the open source movement's approach to be
Law professor and FSF advisor Eben Moglen (one of whose talks I'll discuss
later in this article) said that free software
advocates should let the "gravity of free software" attract more
people to use it. Practical arguments can bring both software
companies and their users further along.
Radiant Data Corporation
Radiant Data are the makers of a distributed, highly available
filesystem called PeerFS. Because it supports Linux, according to VP
of Sales Robert Peverley, the company would like to consider
open-sourcing the system "over beer on a Friday afternoon." I
responded that they'll get somewhere when they start talking about it
Wednesday mornings. But I wasn't rushing him by any means: I
recognized that moving from a proprietary to a partial of full open
source model is a risky decision that has to be supported by a strong
First of all, does anybody want the software? Would anybody take a
second look if it were opened up? There must already be a baker's
dozen free distributed filesystems. Well, Radiant Data thinks they
have something pretty compelling, and others have told them so
too. They are particularly proud of their locking mechanism, which
greatly reduces contention compared to other systems out there.
Second, what tangible benefits would they get by opening part or all
of it? For this question there's a good answer too: they know there
are features they lack that the community might provide (or at least
test and comment on). For instance, all data is currently replicated
in full across all nodes in the system. It might be more efficient to
implement a RAID-5 sort of scheme; they would like to be suitable for
a Wide Area File Services solution.
So we tossed around all the models we knew of and speculated about
whether they'd work for Radiant Data: dual-licensing (a distinct
possibility, because a filesystem is tailor-made for embedded in other
applications), proprietary add-ons, support (not a good option for
them--the system works too well out of the box), parallel free and
commercial offerings, and so on. I'll be staying in touch with
Peverley and report if the company gets to discussing it on Wednesday
Qlusters's SEMPRE product makes it easy for companies to deploy
software on multiple systems for load balancing and high availability;
they claim impressively fast failover times of a few seconds. SEMPRE
controls clusters of Linux systems.
According to Fred Gallagher of Qlusters, the idea of open-sourcing
part of their product came to them when potential customers told them
they had something valuable to offer, but that the customers already
had some part of the solution in place--for instance, using Nagios for
network monitoring. The customers didn't want to throw out the system
that already was working and around which they had built up their
operations. They wanted Qluster to offer an open system they could
plug their existing operations into.
So the first step Qlusters (and Radiant Data, too) need to do is
modularize a fairly monolithic system. Qlusters plans on creating a
lower layer containing such things as their Virtual Environment Domain
(a way to build easy-to-deploy operating systems) and their
provisioner (which lets administrators add and remove systems from the
cluster as needed). They'll provide a next higher layer of hooks, and a
set of useful tools on top. The lowest layer is the one they're
thinking of open-sourcing. In theory, they could provide the hooks
without opening the code, but in practice they think users will find
it much easier and more appealing to adopt their system if it's open
source. They can then offer a proprietary layer of products on top.
A note on clustering
Before I continue with the issue of open sourcing, I'll point out that
Radiant Data and Qlusters both offer types of clustering, and that a
look around the show floor at LinuxWorld suggests it's the most highly
contested market in Linux. It seems that open sourcing may be seen by
companies as a necessity in the fierce race for dominance.
At the Ottawa Linux Symposium three weeks ago (which I reported on
in other blogs),
a proposal was raised to add a field to the task data structure in the
Linux kernel to support clustering. I asked Donna Jeker of
what she thought of this initiative. (The basic idea is to indicate
whether a task is local or remote, because tasks hosted on remote
systems must currently be handled through intricate user-space logic.)
Jeker was noncommittal, saying that they had certainly found their own
solution in user space but would be open to seeing what a kernel
enhancement could do for them.
Dual licensing and MySQL
In my conversations with Radiant Data and Qlusters, the possibility
came up of dual licensing (one open source, one proprietary, for the
same product). The companies best known for their
dual-licensing approach are
(makers of the Qt programming framework)
and MySQL AB.
Dual licensing is a natural for Trolltech, because a programming
library by its very nature is incorporated into the customers code, so
a licensing requirement must be obeyed to derive any use from the
MySQL is different. Millions of sites can deploy it without linking to
its code. The base of paying customers is therefore much smaller,
although many will support the company out of self interest or will
purchase the MySQL Network support package.
The case of MySQL is hard to judge because a couple years ago they
received a massive infusion of both money and programming talent from
SAP, a one-time windfall that alters their funding needs.
MySQL also places tremendous pressure on itself by trying to hire just
about all the talented programmers who know it well and contribute to
it. This is important to their business model because they want to
remain the key innovators for MySQL; a substantial stream of useful
innovations from outside the company would alter their business model.
A firm hand on anarchy: Eben Moglen unfolds (a couple creases of) the GPL 3.0 roadmap
Many of use are wondering what the new version of the GNU General
Public License will address. After Eben Moglen's talk at Linuxworld
yesterday, we're still wondering. But Moglen laid out why Richard
Stallman is proceeding in such a perplexingly closed manner.
What would you do if you had a document that would affect the
computing needs of millions of people in a couple hundred different
countries, many of whom with scaldingly hot opinions of both what you
have now and what you're about come up with? How would you advise
handling an estimated 150,000 individual commentators, and 8,000
Thus the goal of providing a clear beginning and end to the discussion
progress. Moglen could not tell us exactly what is being proposed, but
he did lay out the outline of a process for the proposal:
In October or November of this year, a process document will be
released to "close down uncertainty," so everybody knows how Stallman
The draft will be released at the end of 2005 or beginning of
2006. Discussion will proceed for one year.
In the summer of 2006, a conference will be held in Europe about the
question of translating the GPL. Another such conference may be held
After the year of discussion is finished, Stallman will make the
decision what to put in the GPL.
If this doesn't seem like a particular open-source way of doing
things, it highlights that openness needs structure. There must be a
center somewhere; a place where the buck stops. Given that fixed
point, everything else can be pretty free.
I should mention here that to call Moglen a persuasive speaker is a
tremendous understatement. His passion for what he's doing (he called
himself the "chief facilitator" of the GPL 3.0) is profoundly evident
throughout. The gravity and eloquence of his utterances makes one
feel, even when sitting in a sterile conference room, as if one is
present at the General Assembly of the United Nations. I joked to some
colleagues later that the U.N. should stop fumbling around with ICANN
and try to take over the GPL.
Despite Moglen's' refusal to discuss details, he was willing to mention
some of the general new social and technical trends that have to be
Software patents. While the GPL "can't solve the problem," it may
contribute to an answer. Moglen presented a balanced assessment of the
idea of an open-source patent repository, leaning (in my subjective
opinion) against it. It's worth noting that the OSDL has just
a project to form such as repository.
Web Services (by which he means, I think, application service providers).
Moglen made many other statements about the background of the problem
and their philosophy in handling it, but I will leave those for others
to report on.
Perl over Java
I spent a good deal of time talking to my pals in
makers of an open-source ebook reader called ThoutReader. Writing it
in Java has made it easy to deploy, because apparently the "write
once, run everywhere" philosophy works in this case. But Mark Carey,
CEO, told me that their designer Gary Varnell, together with another
leading programmer in the Perl community, plan to redo the reader in
Perl, achieving a substantial reduction in size and speed-up in
execution. How can this be, I cried.
Well, the main explanation that comes to my mind is that Gary is a
devoted Perl programmer from way back, and that he may do a better job
at coding in Perl. Perhaps he just doesn't grok Java in the same way.
The true coder, like the true revolutionary, is guided by great
feelings of love. (To paraphrase a 60s cliché.) But Carey said
the Perl footprint was tiny compared to all you need to run Java, and
that the Perl modules were easier to structure so as to modularize the
LinuxWorld Expo wrapup
LinuxWorld continues, even as it directs more effort toward
enterprises, to be a significant event in the computing calendar. I
attended many serious and well-prepared talks, and met key people in
the industry ranging from hackers in the .Org Pavilion to vice
The move to Moscone Center West was beneficial. Everything was easier
to find and more logically located. It was nice to have the .Org
Pavilion and other small booths outside of the main hall, away from
the carnival atmosphere, so that one could actually conduct
The control exerted by management was pretty intrusive. Next year,
they can save money and simplify things by RFIDing all of us. Guards
at every door could be replaced by sensors that go off if a warm mass
passes through without a valid RFID. Cameras could monitor our facial
expressions and report our opinions of the speaker more accurately
than evaluation forms. Perhaps while Richard Stallman is circulating
his license for review, IDG can circulate the source code for the
Linuxworld Expo RFID Monitoring system.
Perl over Java has much more to do with preference
Another contrarian blog comment...