Re: [vox] [fwd] flashmobcomputing...
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [vox] [fwd] flashmobcomputing...
Yeah, I was there and it was a blast! I volunteered and was the guy who
introduced speakers in the second track. (I wish I could have been better
prepared for the introductions
They announced this in the gym, and you could read more on the a-frame
chart in the gym if you looked for it, so I do not think this information
was limited to just the press, and the rest is from what I observed as a
The presentation was very well organized, with plenty of helpful people
all working very hard to provide good service and no attitude.
The doors opened early in the morning at around 8:00am and a small crowd
of people came with their computers to set them up. There were *plenty* of
volunteers present and were *very* helpful.
When you came with your computer, you were met at a security controlled
"gate" and escorted directly to a table where other volunteers were
introduced to you and you were handed off to another person.
That person worked with you for a short time to ensure your machine booted
their software and was able to run the test client. Power management in
BIOS was disabled, and after the checklist was completed, you were
permitted to leave or stay and watch.
people continued to trickle into the gym with more computers up until the
last few minutes. (11:00am)
E-loan was nice enough to provide many computers to this test. (Probably
over 100.)(If anyone from e-loan is reading this, THANKS!)
By the time they were getting ready to close of the gym and start
benchmarks, I notice that we only had achieved about 55 to 60% of
capacity. Many tables were empty. (They were well prepared to fill all of
the slots if more people actually came to this event with computers.
At around 11:00am, everyone was kicked out of the gym except for the
volunteers and security guards. Viewing of the machines running in the
large room was limited to a small 40x40 foot space in the gym that was
guarded and two viewing windows outside.
A monitor outside the gym was provided with some virtual data on
processing. Some conflicting information was provided about this monitor.
One or more people said the data on that screen was not "live" but instead
"virtual." I never saw this screen and cannot comment more on it.
Two tracks of presentations (organized by professor Wells) were made
available to people who took part in the event. (They also had a
3d-shoot-em-up competition with Halo and other games going on for people
who did not want the content of the lectures.) (Some of these cost money
with a prize at the end for the winner.)
Both lecture tracks were amazing! They had speakers come from various
companies that provide clustering integration solutions, but sales pitches
were not very common. There was discussion of technology and some of the
people who presented were willing to discuss many of the technical aspects
of how various mods to hardware could impact clustering of systems. They
have a list of the people who came to speak at this event in the two
tracks on their website:
Professor Wells offered a rather interesting historical review of the
system (Colossus) in Bletchley Park used during WWII to crack the various
ciphers used by the Nazis and some information about Flowers, and Turing
but primarily discussed the question: was Colossus a Super Computer?
At around 1:00pm, the first round of tests was completed. They used 256
machines in this initial test and were able to get 177 GFLOPS with a high
water mark of 180 GFLOPS. After this test, they were going to start a test
with 660 machines. Making a number of assumptions (using simple mean
average without consideration of diminishing returns and overhead for
bandwidth and differences in hardware from initial test and memory, etc.)
a simple projected benchmark with all 660 machines would have been at
around 950GFLOPS with 660 machines...
At around 5:30, they ended the second test. It seems at least one node had
problems, and caused the second test to not work out as planned. All that
would need to happen for the second test to fail would be for one (or
more) node in the group of machines to fail. Overheating of CPU,
over-clocking of CPU, bad memory, over-clocking motherboards, etc... any
one of these cause problems, and one of these may have caused problems.
(Based on the level of knowledge demonstrated by the people who
volunteered for this event, I would not attribute any problems with the
second round of benchmarking to any of the volunteers. I would more
quickly assign blame to hardware than anything else.
(Where I was able to get this information:
One of the event organizers and several volunteers in the gym provided me
with the above stats, and the A-Frame update board provided more
information from which I was able to build the above data.)
At around 7:00pm they had a party for volunteers and a LAN party for
others, but I did not stick around for this.
Overall it was a well done event, but it would have been better if I could
have seen some of you there. :-)
Maybe next year!
One of the interesting things I found out about while I was there was
about luserfs: (page that offers some info about it)
Rather nifty :-)
Bill Kendrick said:
> FWD'd from BALUG/SVLUG lists... I knew ME was planning on going. Any
> ----- Forwarded message from Sameer Verma <email@example.com> -----
> Date: Sun, 04 Apr 2004 10:33:21 -0700
> From: Sameer Verma <firstname.lastname@example.org>
> Subject: [Balug-talk] flashmobcomputing...
> I was wondering if anyone from these lists went to the flashmobcomputing
> event. The event itself was an interesting exercise. AFAIK, they made
> provisions for about 1000 machines, but got around 800 or so (these
> numbers are based on what I heard on the floor at the venue). They had
> pools of machines running on a 10/100 switched network, which talked to
> 1 (or maybe 10?) Gigabit switches from Foundry Networks. Each machine
> ran a morphix-based distro with Linpack benchmarking software. The
> distro is available for download at
> They managed to run 256 machines in the end (others were not usable for
> reasons I wasn't very clear about...will have to wait for the press
> release) and peaked at 180 GFLOPS and solved about 75% of the problem
> set (matrix manipulation). In all it was an interesting experience for
> running a cluster off of consumer laptops and desktops (although a pool
> of machines were loaned by e-loan...not exactly grasssroots?).
> Dr. Sameer Verma, Ph.D.
> Asst. Professor of Information Systems
> San Francisco State University
> San Francisco CA 94132 USA
> ----- End forwarded message -----
> vox mailing list
vox mailing list