Linus Torvalds World Tour 99: Balug Chinatown
What's in this page?
After visiting SVLUG last year (in the
Silicon Valley), Linus was kind enough to come to our sister user group, BALUG on Tuesday June 15th, which meets in San
Francisco's Chinatown.
This page gives some info on the little talk he made and the Q&A session
that followed.
Somehow I always manage to write my pages late at night when I'm tired and
even less able to spell or articulate thoughts in a way that makes some sense
to other people, than usual that you have to feel free to point out spelling
mistakes and factual errors if there are any, I'll update the page quickly.
If you are still bored after reading this page, you're more than welcome to
waste even more time by reading other reports I've written, like my pages on
Linuxexpo 99,
LinuxWorld Expo Winter 99, or the various
other reports and pictures I have on my linux page
Marc Merlin
|
|
BALUG
As everyone knows, those who are not fortunate enough to live in the Silicon
Valley, live in San Francisco. While the enlightened ones honor us by driving
down to see us at each SVLUG meeting, they still felt that they had to have
some kind of local linux meeting (as if there was a life beyond SVLUG), and
they felt compelled to create BALUG which meets in San Francisco.
Because they usually are lonely during their meetings, and because at SVLUG,
we care for our fellow linux users, some SVLUG members usually face rush hour
traffic to go to BALUG and make the event more lively.
Oh, and did I tell you about VA and Linuxcare? Mmmmhhh, never mind, forget I
said anything
(Hehe, I'm sure I just made a lot of friends. Man, now I'll have to move before
they TP my apartement and my car
)
Anyway, back to serious business: I have to admit, they handled the large crowd
(close to 400) beautifully, thanks to the great volunteers from BALUG
(under the supervision of Mike Higashi and Dave Mandala, expert event
organisers).
Rick Moen and Nick Moffitt, two of the BALUG volunteers
The Four Seas Restaurant, the restaurant in Chinatown where BALUG meets every
month, and were we enjoy great food before each talk, was really packed that
evening. Say what you want, but Linus moves people.
The presentation
The cool thing with Linus is that he will sign anything you want or also:
books, t-shirts, laptops, you name it (even a leather jacket once). The only
rule he has is no signing of body parts
Dave Sifry from Linuxcare and Chris Dibona from VA Linux Systems made the
introduction, and announced that since Linuxcare and VA were giving money to the
Debian project for each person who attended Linus's talk, about $7000 was going
to be donated. Thanks to both companies.
Linus
Then came time for Linus to speak. Since he prefers interactive talks, he gave
a 10-15mn presentation and then switched to questions and answers.
What he mentionned was that while the 2.3 branch started recently, he definitely
wants that branch to be over much faster than the 2.2 branch which took
forever (2.5 years). What he is looking at is to have 2.4 out before the end
of this year, but time will tell and as he said "I have been known to slip
before"
For now, the two main things for the 2.3 branch, are USB support and performance
enhancements. In the 2.2 branch, SMP was greatly enhanced but there is still
work to do, especially if you look at some benchmarks. It's also a good thing to
improve those for PR reasons.
Linus then switched to Q&A and mentionned the following things:
- On the issue of risks of bloating for fine grained SMP support, Linus
answered that the main problem with OSes like Solaris, is not too much with the
SMP locks, but with other parts of the kernel that were never changed (on
Solaris, the sucky filesystem comes to mind)
Linus is looking at 8, maybe 16 CPUs, but not 256 or 512 CPUs. After 16 CPUs or
so, he prefers to look at clustering because a lot of hardware, especially PC
hardware, wasn't designed for such scalability anyway.
- The usual question of "how about more than 4G on 32bit CPUs?" was asked
and Linus's answer was to use a real CPU (i.e. 64 bits). He however mentionned
after that, that Steven Tweedie is working on handling those machines even
though, according to Linus, the support may not be in for 2.4 and by then, Merced
will probably be out, so what's the point?
As far a large file support on 32 bit architecture, the answer is similar, get
real hardware, and especially the problem is that the huge kludges necessary to
work around hardware limitations are going to come bite us down the road when
the hardware limitations are gone.
- Linus is also interested in reaching the desktop, set top boxes, appliances,
and why not phones... While he doesn't work on it, the desktop is an interesting
universe because you have lots of applications and users who do the most
unexpected things
- For now, he doesn't have a list of what he will allow in 2.4. At this point,
performance and USB are the main focus and we'll see afterwards.
- While the penguin was announced as a 2.0 mascott at the beginning, there's
no plans for a new mascott for linux 3.0
- Some of PCMCIA is going to go in the kernel, but Linus doesn't see all
PCMCIA devices in the kernel.
- He sees some tree fragmentation not for specific hardware architectures,
but for different kinds of uses like desktop versus high end servers and
beowulf machines, it can make sense to have different kernels.
- About linux standards, it makes sense to him to document the existing as
a standard instead of coming up with a brand new thing that people are likely
to cheerfully ignore, especially when it comes in way too late.
- For USB, the existing support wasn't moving very fast, and Linus looked at
it and had a hard time understanding it, so he made the reasonable assumption:
"if I can't understand it, it must be bad"
and while he admitted to not liking having done that (because as a good
programmer, he is lazy that doesn't like to rewrite code), he started a new USB
stack from scratch that isn't as fancy but much easier to understand and he is
happy that some people took over that code.
- At this point, he doesn't see himself as starting a new project, and he
seems to be ready to go at it for another 8.5 years, but after that he wouldn't
mind taking a vacation, if possible on a small island
- For now there is no planned kernel HA support, the idea is not to crash in
the first place. User space solutions can be used instead.
- While we can ignore bad benchmarks, other people don't necessarly ignore
them and someone wrote a really clean implementation of http in the kernel, and
there is a chance for that to go in the kernel.
- About source management, Linus is convinced that CVS is indeed not the right
tool, and he is going to give a serious look at Bitkeeper when it comes out
(scheduled for July 15th). (Bitkeeper is from Larry McVoy, and you can see
my report of his talk at
linuxexpo if you'd like more info about it)
- And last, but not least, he told us what transmeta does: Cool stuff
Misc
Linuxcare did it again, this time they used Debian for their Simply Palm-like
poster ad.
Mike Higashi showing off the new Linuxcare poster
See this bit of my linuxexpo
report for the Red Hat version of this poster, and the story associated
with it.
If you want free palm pilots, write your own free OS
This time, it's linuxcare who gave Linus a Palm Pilot, version V, no less...
Max's
Because the real linux geek has no life outside of CABAL meetings (there is no
CABAL), we couldn't quite go home yet, and because you can actually find places
to eat after 22:00 in San Francisco (other than Denny's that is), we went
to Max's
Thanks
Thank you very much to all the fine folks from Linuxcare, BALUG and the
other volunteers who made all this happen and to Sam Ockman for getting
Linus as a speaker in the first place.
If you want to see more pictures I took that evening, you can browse the
picture library. You can also find an article from
Reuters
here
See y'all next time.
Marc Merlin
(marc_web@merlins.org)
Email
Link to Home Page
99/06/17 (01:48): Version 1.0
99/06/17 (09:15): Version 1.1. Added the date of the event
99/06/17 (12:25): Version 1.2. Fixed a few references from Linuxcare to
BALUG. Credited Dave Mandala
99/06/17 (15:33): Version 1.3. Fixed a few spelling mistakes that were reported
to me
99/06/23 (23:10): Version 1.4. Added link to article on Yahoo