Ask the All Purpose Fox - Enterprise!

Started by Kenji, January 18, 2008, 09:45:27 PM

Previous topic - Next topic

ShadesFox

Maybe.  Maybe I'll just chew on it.  NOM

That is a lot like asking, "Except for everything, what is the use of anything?"
The All Purpose Fox

Azlan

I must note, no one has yet survived the entrance exams to APFU... I'm glad they pay up front and all the scholarships are fully funded for the term.
"Ha ha! The fun has been doubled!"

Darkdragon

But I thought it was free if you can get in... Was I mistaken?

ShadesFox

We tell everyone that it is free.  We send everyone an invoice as well.  It is full of amusing.
The All Purpose Fox

Darkdragon

Ah... Do you mean something akin to "Hidden charges"?

llearch n'n'daCorna

Not really. It's not like they're hiding them.

You don't have to pay the invoice. It's just a bonus if you do.

Isn't that a clever trick? They've got more of them. I think they've got more of them. Haven't they?
Thanks for all the images | Unofficial DMFA IRC server
"We found Scientology!" -- The Bad Idea Bears

Darkdragon

May I ask the APF's to share some of these neat tricks?

Azlan

"Ha ha! The fun has been doubled!"

Mao


Roureem Egas


Myxini

I have a 2 pound chunk of daikon in my fridge, what should I do with it?

where did I leave my spool of black silk thread?

How can I learn to stop talking about fascinating-but-generally-perceived-as-disgusting things when eating?

Gabi

Quote from: Myxini on October 27, 2008, 05:25:04 PM
I have a 2 pound chunk of daikon in my fridge, what should I do with it?

where did I leave my spool of black silk thread?

How can I learn to stop talking about fascinating-but-generally-perceived-as-disgusting things when eating?
It's up to you. You can eat it, do something creative with it, or leave it where it is.

At the last place where you had it.

Easy: keep your mouth shut.
~~ Gabi a.k.a. Gliynn Starseed, APF ~~
Thanks to Silver for the yappities, and to everyone for being so great!
(12:28:12) llearch: Gabi is equal-opportunity friendly

Slowtini

You see an individual, he is wearing a hockey mask, carrying a chainsaw, and also has a bloodied apron about his waist.

What do you do?

ShadesFox

Quote from: Slowtini on October 31, 2008, 10:31:38 AM
You see an individual, he is wearing a hockey mask, carrying a chainsaw, and also has a bloodied apron about his waist.

What do you do?

Shoot him in the face.
The All Purpose Fox

VSMIT

What if you don't have a projectile weapon?

Roureem Egas

Quote from: VSMIT on October 31, 2008, 12:46:47 PM
What if you don't have a projectile weapon?

Wield a folding chair for crazy awesomeness.

Corgatha Taldorthar

Instead of sacrificing my own soul for wealth or talent or supernatural aid, I was wondering how I could sacrifice someone else's soul for the same benefits.
Someday, when we look back on this, we'll both laugh nervously and change the subject. More is good. All is better.

Kenji

Quote from: Corgatha Taldorthar on November 04, 2008, 09:10:00 AM
Instead of sacrificing my own soul for wealth or talent or supernatural aid, I was wondering how I could sacrifice someone else's soul for the same benefits.

Sure, but the police won't like it.

Gabi

No questions! I'll have to ask one, then.

Are the typable terms in the combinatory logic with patterns (using the type system with pattern-decorated arrows) strongly normalizing?
~~ Gabi a.k.a. Gliynn Starseed, APF ~~
Thanks to Silver for the yappities, and to everyone for being so great!
(12:28:12) llearch: Gabi is equal-opportunity friendly

Kenji

Quote from: Gabi on November 10, 2008, 06:48:57 AM
No questions! I'll have to ask one, then.

Are the typable terms in the combinatory logic with patterns (using the type system with pattern-decorated arrows) strongly normalizing?

I'm sorry, but all patterns besides pinstripes have been deemed legally offensive on 4 continents.

Corgatha Taldorthar

Someday, when we look back on this, we'll both laugh nervously and change the subject. More is good. All is better.

Darkdragon


llearch n'n'daCorna

Quote from: Darkdragon on November 10, 2008, 05:28:37 PM
What is the best name for a pet?

Lunch.

Dinner is passée, and you really don't want to get into the dog breakfast...
Thanks for all the images | Unofficial DMFA IRC server
"We found Scientology!" -- The Bad Idea Bears

ShadesFox

Quote from: Corgatha Taldorthar on November 10, 2008, 02:12:30 PM
Beezer or Zeji?

Benny.

Quote from: Darkdragon on November 10, 2008, 05:28:37 PM
What is the best name for a pet?

Llearch

Quote from: llearch n'n'daCorna on November 10, 2008, 06:05:53 PM
Quote from: Darkdragon on November 10, 2008, 05:28:37 PM
What is the best name for a pet?

Lunch.

Dinner is passée, and you really don't want to get into the dog breakfast...

Same thing really.

Now for my questions:

Boston or Chicago?

And if you were to optimize a program running dense linear algebra and heavy with matrix transformations on a 60 node cluster, each node with a dual core processor and connected with some really sweet interconnects, would you be better attack the problem of memory latency or network latency?
The All Purpose Fox

Slowtini

Let's say, hypothetically, someone were to come across a way to screw with the fundamental levels of reality, and that hypothetically, they could do so at will.

What would be the best thing to do, in this completely hypothetical situation?

llearch n'n'daCorna

Quote from: ShadesFox on November 10, 2008, 10:43:56 PM
Boston or Chicago?

And if you were to optimize a program running dense linear algebra and heavy with matrix transformations on a 60 node cluster, each node with a dual core processor and connected with some really sweet interconnects, would you be better attack the problem of memory latency or network latency?

New York.

Depends. If you have sufficient memory on each node, then network is probably your biggest bang/buck. On the other hand, if you haven't got enough memory, all the network efficiency in the world isn't going to make things any faster, and screwing with the standard protocols is asking for interesting bugs to show up.

Are you using NUMA on the cores, or are we talking standard Wintel hardware?
Thanks for all the images | Unofficial DMFA IRC server
"We found Scientology!" -- The Bad Idea Bears

ShadesFox

#986
This is standard Wintel hardware, each node had 4 gigs of memory and a dual core Xeon processor.  Actually it is a Dell blade system.  BTW for anyone looking at buying a cluster from Dell... don't get the support package if you do.  Of any sort.  What ever they tell you about support and delivering the system preconfigured is a lie.

Thing with the code is that it looks like there has been some optimization for the network transfer latency already while there has been nothing done about memory access.  And some profiling seems to indicate that the matrix transformations chew up time at parts that have nothing to do with network transmissions.  Though that is just a small profile run that did not hit that many nodes so the network latency may grow faster then the memory latency.  I still need to run more tests.

Quote from: Slowtini on November 11, 2008, 02:12:39 AM
Let's say, hypothetically, someone were to come across a way to screw with the fundamental levels of reality, and that hypothetically, they could do so at will.

What would be the best thing to do, in this completely hypothetical situation?

Have God visit Richard Dawkins.
The All Purpose Fox

llearch n'n'daCorna

Quote from: ShadesFox on November 11, 2008, 12:45:49 PM
This is standard Wintel hardware, each node had 4 gigs of memory and a dual core Xeon processor.  Actually it is a Dell blade system.  BTW for anyone looking at buying a cluster from Dell... don't get the support package if you do.  Of any sort.  What ever they tell you about support and delivering the system preconfigured is a lie.

"enough" depends on what you want to put in it. 4Gb may be short. 256Mb may be ample. Sadly. :-/

As for Dell, I don't trust salesweasels to give me the time correctly. Particularly if there's money in it for them.

Quote from: ShadesFox on November 11, 2008, 12:45:49 PM
Thing with the code is that it looks like there has been some optimization for the network transfer latency already while there has been nothing done about memory access.  And some profiling seems to indicate that the matrix transformations chew up time at parts that have nothing to do with network transmissions.  Though that is just a small profile run that did not hit that many nodes so the network latency may grow faster then the memory latency.  I still need to run more tests.

That's usually the best way to figure these things out.

Feel free to discuss it with me, although I probably don't know as much as you about it, and may not be able to help much...
Thanks for all the images | Unofficial DMFA IRC server
"We found Scientology!" -- The Bad Idea Bears

ShadesFox

Quote from: llearch n'n'daCorna on November 11, 2008, 01:31:42 PM
That's usually the best way to figure these things out.

Feel free to discuss it with me, although I probably don't know as much as you about it, and may not be able to help much...

Well, moving from a cluster with 4 times as much memory per node to a cluster with more and faster CPUs improved the speed significantly on a node per node basis, so there is enough memory.  It comes down to a problem of access time.

And I find that talking with people who understand the fundamentals of computers often yields good advice, regardless of where they actually specialize ;p

And if I do get some really useful stuff talking from you I think it would be amusing to put into a thesis acknowledgement section, "And a thankyou to (insert llearch's real name here) who is a veritable box full of good ideas."
The All Purpose Fox

Zedd

A baby weighs 5 pounds and a cannon ball weighs 7 pounds...Which one will reach the ground first on top of a house and probably bounce first?

Why do revolving doors exist anyhow?

Is it better write on a full stomach or an empty one?

What's the worse wake up call you have gotten in your life?