?

Log in

No account? Create an account

Thoughts while traveling, feeding the rat....

« previous entry | next entry »
Sep. 12th, 2007 | 08:21 pm

I'm on a flight from Japan to South Korea.

The woman next to me is applying more makeup then the average clown
wears... thankfully she stopped spraying on perfume and switched to
bathing in it.

Yuck.

You can just see the wrinkles... got to love the effect.

I'm reading "Feeding the Rat" which is an awesome book. Its making me
want to get off the plane, and go find some place to climb...
unlikely in the extreme but its still a nice thought.

I would really settle for hacking, but the laptop I am carrying
barely lasts a hour and a half on the current battery. Which is
unfortunate since I have several projects I am working on... which is
always the case.

The book has me thinking a bit about the main character and his life.
"Mo" works to climb. I can respect that... I am no where near the rat
that he is. As a climber I enjoy it a great deal, but I don't truly
jones for it, and as a risk taker I am quite the wimp. I rarely take
risks while climbing, and to me the fun is the puzzle. I am terrified
of heights, and I hate to fall. I push myself, but only moderately.
Its like running for me, but without the "knee is going to fail any
moment" that I get when running.

Hacking is a bit different. I do it because I enjoy it, and I
continue to enjoy it every day more and more. In undergraduate I was
once asked "why don't you just do computers and skip out on the
environmental stuff". My answer was "I really only like computers
when they are networked together, and for that you have to be in
academia..."

You can date me based on that statement.

In graduate school I became more interested in networked computers
because I like to look at how to solve problems in parallel. I really
enjoy working on problems which can be solved by reduction. In
graduate school I spent my time looking at watershed models. I became
to really not care so much about the model itself, and more about how
to solve the problem in parallel. The model I designed really was
just a giant bathtub in the end, I got tired of dealing with the
uninteresting problem of terrain. The interesting part was solving
the problem of having many threads work on a particular map at once.

PVM was to me at the time a pretty sharp idea, but I didn't care for
the latency in the solution. Combining computers together was fun,
but I didn't like how slow it was.

Getting access to a four processor SGI ONYX was nice. I spent a
number of weekends compiling and recompiling to solve problems on it.
The lab I had it in for a while had only enough power to either have
an air conditioner or a running ONYX. I bathed in the heat and
skipped the cooling issues (which would surprise no one who has spent
anytime in an office with me). I adore heat. Give me an office at 92
degrees and I am as comfortable as comfortable can be. No light or
humidity though, just warmth.

To me hacking is only getting more interesting. More processors allow
for more threads which in turn makes parallel solutions interesting.
One of my spare 32 some odd projects right now is re-implementing the
Intel C++ threading libraries. I started working on it while at
OSCON. Mostly for academic reasons, but I've found now a couple of
places where I can make use of it. At this point I can't find any
interest in projects which don't have some element of parallelism to
them. I believe that all of the tools I find today, and languages,
to work on these problems are woefully inadequate.

The T-Shirt that I saw recently "life is too short to debug threads"
has me thinking that we need to solve this problem. Whether you are
working on issues of parallelism on a single multiple CPU machine, or
dealing with a cluster of computers, the solutions are really the same.

Calling sleep() is no longer acceptable for timing solutions (and
while I rail against this... I know I have done it as well). Much of
the literature in academia on this subject needs to be dragged into
the the daylight and made available in a clear manner.

The plane is descending, the stow your laptop message is being
announced... and the woman is still putting on makeup...

Link | Leave a comment | Share

Comments {9}

Lover of Ideas

(no subject)

from: omnifarious
date: Sep. 13th, 2007 07:12 pm (UTC)
Link

The T-Shirt that I saw recently "life is too short to debug threads" has me thinking that we need to solve this problem. Whether you are working on issues of parallelism on a single multiple CPU machine, or dealing with a cluster of computers, the solutions are really the same.

I've thought this too. They are the same in so many ways. Local context vs. remote for example even makes sense in shared memory multiprocessor systems because of caches.

I've always thought the best overall model to use was of many independent entities that share as little state as possible communicating through messages, which are essentially pieces of state that move from place to place.

Reply | Thread

Jayson Obedoza

(no subject)

from: jaobedoza
date: Mar. 1st, 2008 03:56 pm (UTC)
Link

dont you just hate it when you find yourself sitting beside someone like that on a public transport!

Reply | Parent | Thread

Brian "Krow" Aker

(no subject)

from: krow
date: Mar. 1st, 2008 04:22 pm (UTC)
Link

Independent entities moving in a shared transport?

Parallel communication is really limited to near CPU at this point. Everything else is locked/semi-locked transport lines with ever increasing message handling.

We don't add lanes, we make the pipe bigger.

The original parallel port, if I remember correctly, had 3 wires for receiving and five for sending. Serial was well... 2 pins (takes four to make it all work unless you screw with the ground).

Look up HIPPI sometime :)

Reply | Parent | Thread

Lover of Ideas

(no subject)

from: omnifarious
date: Mar. 1st, 2008 05:15 pm (UTC)
Link

HIPPI was the original Infiniband, except it used a LOT more wires. I think it's funny how most things that use lots of wires nowadays use them in groups of 4 as decoupled serial links whereas even 5 years ago they mostly carried tightly synchronized signals in parallel.


I should've been able to see that it would happen 15 years ago when I heard the stories about how people putting together Crays had all these wires carefully cut to very precise lengths to make sure it all hung together. But I didn't put 2+2 together.

Reply | Parent | Thread

Brian "Krow" Aker

(no subject)

from: krow
date: Mar. 1st, 2008 05:31 pm (UTC)
Link

Yep... though the problem is with the "lots of independent serial links" is that you can not put enough of them together yet :)

For me...

Though, honestly the bigger problem is getting a computer that can keep up in general :(

Reply | Parent | Thread

Lover of Ideas

On a side note, spam

from: omnifarious
date: Mar. 1st, 2008 05:44 pm (UTC)
Link

I looked at jaobedoza's journal. It's clear from reading it that he's basically going around making comments that seem vaguely topical in some way in the journals of people who have huge friends lists or otherwise have lots of Google juice in order to pump up his journal by the backlink to it.

Reply | Parent | Thread

Brian "Krow" Aker

Re: On a side note, spam

from: krow
date: Mar. 1st, 2008 06:09 pm (UTC)
Link

Ha!

I searched the HTML in his comment to see if it was SPAM. I was wondering about the post... oh well... the comments make it worth keeping :)

Reply | Parent | Thread

Jayson Obedoza

Re: On a side note, spam

from: jaobedoza
date: Mar. 2nd, 2008 11:06 am (UTC)
Link

my comment was

"dont you just hate it when you find yourself sitting beside someone like that on a public transport!"

i was referring to this part

"I'm on a flight from Japan to South Korea.

The woman next to me is applying more makeup then the average clown wears... thankfully she stopped spraying on perfume and switched to bathing in it.

Yuck.

You can just see the wrinkles... got to love the effect"


as i have experienced this more than one time while on a train.


and yes i am trying to get my journal indexed by making a valid comment

Reply | Parent | Thread

Brian "Krow" Aker

Re: On a side note, spam

from: krow
date: Mar. 2nd, 2008 02:08 pm (UTC)
Link

Well, you are being honest about it :)

Reply | Parent | Thread