Stefan's secret pictures

This album is far from being complete, there are some missing pictures from Electronica/98, Linux-Expo Paris/99 and Comdex Chicago/99.


My pets

My name is "Pezi"
I'm a fighter.
My name is "Buggy".
I like all cats.
My name is "Petra".
I can clean the dishes.
My name is "Chico".
I'm too old for having sex.
That's me.
I generally trap into problems.

Comdex, Las Vegas 11/98

Comdex in Las Vegas (most Americans think that this is the
lagest computer show in the world)
was stupid.
My first arrival in the States was not very impressive.
Stupid show, except that I met old friends like Maddog or Jeff Carr (PPC).
There are no pictures available here and I'm not willing to
post here some pictures made at Disney-Land.
Walking arround there with my Penguin-Pullover caused some people
to ask which Disney story I'm comming from.

Long discussions about Linux and after that when exchanging business
cards I recognized that the last round of beer was payed by Ran McNab
"Vice President and General Manager" of Corel Computer.

Linux Cluster, Paderborn 12/98

The CLOWN (CLuster Of Working Nodes) project in Paderborn had the
goal to set up a cluster made of more than 512 Linux machines within
two days.
It was a heterogenous cluster with over 450 intel machines
and 51 666MHz Alpha 21164 machines. I was responsible for
the Alphas.
This huge project was broadcasted by WDR live, eight hours of
Linux and it was the second largest project of WDR in 1998
just the soccer WM was a bigger project for them.
Sebastian Hetze and I build Linpack on the cluster, time was
short so we got only 28 GFlops out of the Alpha machines.
Later at Heise-Verlag I was able to tune the configuration a
bit and I first hit the 30GFlop barrier and later I reached
31 GFlops.
The special thing with the Alpha-Cluster: 50 diskless
and headless nodes driven from a single boot-node
(bootp/tftp) that had two software-mirrored Disks
beside 1GB RAM and six ethernet cards.
To avoid nfs-rooting I ported initrd to Alpha and
a special unique feature is that booting a bootp-kernel
with attached ramdisk now works.
This makes the diskless machines independend from the
boot server after booting, the stuff is now in the
standard Linux-Kernel distribution.
I really want to thank Quant-X and Samsung for supporting
the project within 5 days, making me to the project leader
of the "Samsung and Quant-X Cluster Project" and putting
about 770K USD into this.
Configuring the 50 node 666MHz Alpha clusterChecking the cluster
Integrating the Alpha cluster into the whole cluster
Marc Lehmann beside me
A small overview
Another viewGood organization
Thanks to the fire departement of Paderborn
My alphas need cooling
Many visitors
Done, ready for the Guinnes book Happy people

Heise Verlag, Hannover 12/98

The Alpha-Cluster of the CLOWN (CLuster Of Working Nodes) project in Paderborn was directly shipped to Hannover for some testing in the test-labs of Heise. I was flying to Hannover (having 25GB RAM in my luggage) to build the cluster up again and to speed things a bit up. My plan was to stay there one or two days but it was so interresting that I was there a whole week. Heise-Verlag is very programmer friendly environment, Coffee for free and a cigarette machine. To let the cigarette machine get warm we disabled the "Rauchmelder" (not in my dictionary, sorry) in the test-lab.

There is a testreport in iX 2/99, a good joke was the "high competence of the vendor" mentioned in the summary of the article. A nice joke which I tread very personal. ;-}
Penguin indicator.
We started arround 212 of the Top-500 ranking
Linpack: a bit below 30Gigaflops
We only did serious work and research.
Penguins are very cool.
Ralf Hülsenbusch (manager of the test-lab) was happy
because I built up 30 (out of 50) machines allone.
Doesn't this look nice ?
51 machines in 17x3 setup
typical: My Omnibook on the left side, my cigarettes on the right.
Linpack Table (created by GNUplot)
Created by parsing the linpack results directly
It looks much better in the iX-article.
After a week of hard work:
More than 31 GigaFlops, Top-500 ranking: 202
Because we put much brain into optimizing the cluster
it was nearly 10% faster than in Paderborn.
Nearly every hour we bet ourselves and we reported
the results to national (Germany) and international press.
Fact: We were faster than a Compaq Cluster that costs more
then 10 times than our configuration.

CeBIT, Hannover 3/99

CeBIT was hard work for me.
I was responsible for a 30 node cluster at Samsung,
a 5 Node-Cluster (first showed at Elektronika 98
in Munic) at Bee, and the Linux International booth.
I did also the setup for all machines on the RedHat booth.
I was happy to met Jon "Maddog" Hall again, really a good friend
of mine. We managed the Linux-International both together with Dan York (Linuxcare).
Jon, Dan und ich 3 Pinguine
No, we did the stuff at the both with two additional peoples. Old friends from Paderborn (Marc Lehmann and Lukas) helped us a lot. Marc is the initiator of the EGCS-Steering Commitee and the real reason for the merge of gcc and egcs. He is the author of pgcc, a pentium optimized gcc variant. Maybe he is the only person on this planet who is able to talk to Richard Stallmann in a normal way, but even Marc has his problems with rms. Marc is also a Perl expert, he is writing a book for O'Reilly about Perl modules. His Gimp-Perl module is one of the best peaces of software ever written. Because marc is genius, he got an Alpha machine from Samsung and Quant-X.
left to right: Lukas, Stefan, Marc Flooded by many peoples.
Massenauflauf Es gibt auch unbrauchbare Software auf Linux

Comdex, Chicago 4/99

The amercan way: Bad show. I visited with a small group of people (Rüdiger Örtel, Larry Augustin, Jon "maddog" Hall, Dan Yocum, Matthew Cunningham and Linus Torvalds) the famous Fermilab in Batavia, Illinois. There is a cool report (with nice pictures) about that in Linux Journal 7/99. Dr. G.P. Yeh - he discovered 1994 "top quark" - guided us through the large area.
It's very easy to understand the calculations at Fermilab. A 100 Ton magnet with some sensors costs arround 100 Million Dollar. Later on this day we went for pizza at Jeff Gerhardt's home and Linus - even after some beers - did not tell us exactly whats his job at Transmeta. Something related to wireless communication, top secret.

2-Sep-99, last update 1-Nov-99 Stefan Traby
$Id: english.htm,v 1.2 2000/01/14 02:31:47 stefan Exp $