2001.12_Alan Cox Interview.pdf

INTERVIEW

NUT- CRACKER

ALAN COX

Linux Magazine

talked to him about

his views on kernel

development and

what might be in

store in the future

Linux Magazine What drives you to do all this

good work?

Alan Cox I enjoy it!

LM How do you find the time to do it all?

AC Large amounts of sleep. I work US hours so

sometimes I don’t get up until midday.

LM What do you consider to be the most vital

pieces of software that are missing from Linux?

AC Better calendaring software, certain

groupware programs. The big ones are now

starting to fall into place – office suites like

StarOffice and the KOfficework. Especially with

StarOffice going to OpenOffice. The Ximian guys

are working with Evolution to create a complete

clone of Outlook with all the features and then

some. A pure mailer program is Sylpheed – It

means “Wind Spirit” in Japanese.

LM Are the software support models for

companies correctly set up?

AC Support costs real money. You can pay large

amounts of money for complete enterprise-wide

support or just the back-end part. It depends on

what you negotiate. All the support is there in theory.

The Red Hat guys think they are doing a wonderful

job but you should really ask the customers.

LM Development of the kernel does not use the

CVS model. Why not?

AC The kernel proper does not use CVS but some

developers use some for their parts. The big problem

with CVS is that it is not a good way for a single

person to have an overview of everything going in

and the right kind of quality control and auditing that

we require. I believe that Linus is using CVS as he

wants to see everything in his CVS tree.

LM How does the kernel grow and develop?

AC A directed explosion is the best answer.

Development goes off at all kinds of angles from a

huge number of people for a large number of

reasons. Sometimes it is because they see a financial

advantage – if we pay someone to do this, then we

can sell that. Other times people do it because they

like a piece of hardware or they have bought a USB

device that is not supported and think that “this is

cool and I want to use it under Linux”. Some people

do it out of academic interest, researching a given

topic to improve a piece of software.

LM Does it come about that features are left

hanging because everyone wants the feature

but no one gets around to doing it? How is this

co-ordinated?

AC It’s not coordinated. It does sometimes happen,

but eventually it irritates someone enough that they

fix the problem. It is not uncommon that we have a

piece of hardware that someone has written a driver

for, but it is not really being maintained or the person

who wrote it did not care about certain machines

that it did not work on. If there are enough people

using it then someone will sit down and say “okay I

need to fix this” and then do it. Sometimes it’s the

vendors, because when they run their QA test there

is a problem. Often it is the end users. For example,

those people with very old Soundblaster CD drives, if

we break something then they still fix the driver.

LM Do you get much in the way of requests

from users who are unable to contribute directly

to the kernel development?

AC We certainly get feedback saying it would be nice

if Linux did this or if Linux did that. The vendors are

very good at getting feedback - “we would install five

hundred machines but...”. It can be very useful.

LM Do you think that there are good lines of

communication from end users to coders?

AC Yes. To the extent that what matters is that the

end user is prepared to do the work or that they are

prepared to pay someone to do the work, and that is

how things come around.

LM With the continued growth of Linux, do you

see any downsides?

AC Having a larger mailbox is the biggest potential

downside. In the early days the Linux kernel would

get two or three messages a day. It has continued

to grow with more people becoming involved and

more happening in the kernel. I do not know how

much longer that will continue to happen -

essentially we are running out of things that are

important to add to the kernel. Most of the really

critical things are now in user space. For example,

the world of KDE and GNOME, 3D graphics and all

those kind of things.

LM What is in the next version of the Linux

kernel? Is that the type of question that’s even

polite to ask?

AC It’s the type of question we don’t know the

answer to! There are things people are contributing

which look like obvious candidates: An improved

input layer; the ability to have multiple monitors and

LINUX MAGAZINE

Issue 15 • 2001

INTERVIEW

multiple consoles used more sensibly; plug two mice,

two keyboards and two monitors into one Linux box;

various pieces of filesystem stuff – XFS, JFS. Compaq

are donating clustering code, which is a very

interesting and exciting area. You would be able to

have a room full of Linux boxes acting as one system,

but on top of that, if you lose specific machines then

the system continues without a glitch. This is very

important for a lot of business applications.

LM Filesystems, then. Do you have a favourite

filesystem?

AC Journaled filesystems are useful for lots of

applications. It really depends on what kind of thing

we’re trying to do. So, we have things like ext2

which migrated into ext3 – a journaled filesystem,

which does exactly what every standard generic

filesystems has done over a period of time. Reiser FS

has done a lot of work on small files, on faster

directory handling. IBM’s JFS is looking extremely fast

– it is interesting for that. We will see, I think, over

time, which one will work out for the best. It’s a bit

muddy at the moment for some of them.

LM In what way will that muddy water clear?

Who will be the victor?

AC I don’t actually think there will be a victor.

Before, we had competing filesystems – ext fs, xiafs

– that was a long time ago, then ext2 came along.

Pretty much everyone ended up running the same

filesystem after six to nine months, simply because it

was the natural one to use. It was the one everyone

else used and it seemed to work. I think the vendors

will ship the filesystems that work. They do a lot of

QA testing on that. In some cases we have specialist

filesystems: JFFS for Flash devices is very useful for

the iPAQ, but completely useless for your average

hard disk user.

LM What hardware do you recommend and do

you get involved with hardware concerns?

AC I try to pick devices with free drivers which work.

Like everyone else, I’m trying to build myself a

machine that works at a sensible speed or uses as

little power of whatever. I play with a fair amount of

stuff. I build machines that are all Athlon because

that gives me the best build performance. For

desktop machines I’ve been playing with the new VIA

C3 chip – it doesn’t need a fan and is so much

quieter, but it’s not a speed demon. I’ve not really

played with the Pentium 4, some benchmarks but

that’s as far as I’ve got. It’s the first generation of the

chip and I think the real question is not how the P4

performs now but in twelve months time.

LM The support from hardware manufacturers is

growing but do you think anything can be done

to encourage them more?

AC Most of them make the openness decision based

on business risk, or financial reward. So, for example,

a lot of small USB vendors have no secrets. Some

hardware vendors like SIS, who are working very

closely with the Linux community, are keen to give

good support. We also have people like nVidia who

are more worried about not giving away secrets to

rivals. They are worried that if they release their

software technology then someone is going to use it

and make the ATI Radeon run even faster than their

card. You talk to these people and sometimes it

makes sense, sometimes it doesn’t, but at least you

understand their point of view.

The other thing that has really helped is when

people like Compaq and particularly Dell and IBM get

involved because when they are building a server

they think Linux is going to be one of the supported

systems. It means that they go to the hardware

vendor they buy from and say “If you have Linux

support for this board then we will buy them in large

numbers and sell them in our servers”. For many

more conservative businesses the fact that you can

actually say: “Well, if we will do this piece of work for

this approximate cost, we will get this money back”.

It obviously makes it easier for them. Many of them

are generally uneasy about giving out documentation

to you, as they are more used to a traditional

business relationship.

LM We were talking to someone at the Systems

show, they offer the service of a manager/coder

interface. Is this common?

AC It’s one of the jobs that LinuxCare have been

doing. It is to a certain extent, part of what Red Hat

offer in doing direct device driver work. We will write

you a driver, we will help you commit it to the

mainstream kernel, if that’s the way you want to go,

and we will convert the things the techies say to the

things the management need to hear about “time

scales and pricing”. It’s no good asking your average

programmer, “How long will this driver take?”, the

answer is always far too short, so the project will

overrun, and they are often far too vague.

LM Do you use a desktop or are you a

console man?

AC Well, I mostly use X, I tend to use XFce for my

desktop most of the time, then run mostly GNOME

applications, sometimes KDE. You can set both

GNOME and KDE up to look the same. In the

desktop world, I am very much an end user and as

far as I am concerned it all looks the same. OK, some

bits are QT, some bits are GTK, but, who cares?

LM You have a very good line of communication

to the community, tell us more about your diary.

AC Well, the diary was originally set up pretty much

for that purpose because when I joined Red Hat one

of the things Eric wanted, as my then manager, was

a monthly summary of what I was doing. So I

figured out seeing as this was open source the

monthly summary probably ought to be, so that

became the diary.

It’s no

good asking

your

average

programmer

, “How long

will this

driver

take?”, the

answer is

always

far too

short

’

Issue 15 • 2001

LINUX MAGAZINE

‘

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: