Wednesday, March 16, 2011

Little High Little Low


The 'low' of today was the devastating video of the tsunami flooding, the flood of flu in which I am still woefully drowned, and the MS class which was held today, because of the upcoming match on Saturday (it was originally scheduled for Saturday). Like we're going to win, but you never know. Dil jalaain to sub saath mil kay jalaain, right.

Anyway, the class today :(. Lemme walk you through it. [why? You may wonder.. simply so ... 'exclaim karain to saath mil kay karain'] :'(

So it had all started off rather innocently. Advanced Parallel Programming? Interesting! I was so excited. Maybe we're finally going to learn programming of real parallel systems. Me and my excited-about-coding thoughts.

None of that happened.

What happened instead was... that we're stuck (for 3 lectures now) on Communication Models and Implementations... which essentially means that if you have a quad-core machine, the four processors nestled inside your motherboard obviously can't stay still. They have got to jabber to each other. And jabber they will - in peculiar patterns. Patterns YOU decide. Jabber they will... at particular times.... times YOU decide. Jabber they will - in peculiar ways... ways YOU come up with.

So let your imagination run wild and come up with horrific ways of having your processors chatter (I mean communicate) with each other. And you'll come up with these kinda imagery.

And if you're ever nerdy enough to think that hey, this looks easy ...two dimensions yay.

So ha! Take a look at this. Your multi-core system might be sustaining processors connected in THIS way:

And the types of communication operations...left us all baffled.

Sir started off with a super simple example of a broadcast - e.g a radio - one sender .. and multiple receivers. Eaassy peaasy.

The easiness peasiness disappears the moment you begin learning about: one to all broadcast - all to one reduction.... all to all broadcast... all to all reduction.... all to all personalized broadcast....all to all personalized reduction......and it goes on and on! All of them are very subtly different from one another. 

My mind was in a circular shift, by the end of the show. I mean, class. Even the teacher looked exhausted. And complained that it's 'lack of interest' that's making our faces 'long' ...

I am fascinated by the challenge. Laikin samajh to ayy!! :@

Oh, yeah. Saw the title of the post and remembered the 'high' - 
Finally got my jee rrr eee account unlocked and saw my percentiles 

:D. 98 in verbal and 94 in quantitative. And a 5.0 in analytical (er, highest is 6 so this could have been improved sure). 

But anyway, Alhamdulillah :)


Tauqeer said...

Easy sa to hai. lol

majworld said...

i waited 5mins to first revise my comp archi course in mind nd think on this quad core processor problem..its really exciting..i guess wat we read there was how to reduce time of executing instruction set with in a single single cycle, multiple cycle nd i never thought on this before..comm btw multiple with quad core isn't the communication easy as there r 4 just 4 nodes?? or are there processes inside a processor contending for communication?? nd why this 3d pattern needs to be built..i mean y we need to imagine this problem in 3d when it could be done with simple graph theory or other very simple protocols :S..nd which programming will b used for this parallel coding?? like u people wud use HDL language or some other??
sry for the long comment..its really exciting for me, dats why excited to know:p

congrats on the complete gre result..:)

Uni said...

I wish you would see the intricacies involved in this kind of stuff before making such a nice statement.. *scowl*

It sure is exciting.. lol. But very tough. 4 nodes seems easy, yes. But when a hypercube is considered, it's actually 8 nodes.. (what we were doing in class).

Also, there are 32-core machines available (in academia) today. 64 are coming up. Things get WAY trickier!

Yes, there are cores within physical processors. But we were considering (in class) nodes (processors) only. Not cores.

The problem isn't imagined in 3D. The communication pattern is such, that it's either done in a linear array (simplest), or a mesh (2D) .. or a hypercube (3D)... and hypercube is the more realistic scenario.

Programming :( .. Sir is showing us algorithms... (not code). And the library used practically is MPI (message passing interface). No, not HDL.

It's exciting yeah :).. But I find real coding more exciting. This stuff is all theoretical and kind of difficult to visualize..

Thanks very much!

Thanks both for dropping by!

Uni said...

But when a hypercube is considered, it's actually 8 nodes.. (what we were doing in class).

Er, a hypercube isn't necessarily comprised of 8 nodes.. I meant a 3-dimensional hypercube here.

Otherwise, the dimensions go up and so do the no. of nodes.

Tauqeer said...


Tazeen said...

*Blinks* I can't believe I read that. But the picture's cute so that's good and congratulations on finding your password! :D

majworld said...

oh ok, the 3d is used for computations with arrays..i ws surprised that its only time contention between nodes and not spatial contention that we considering 3d. anyways seems tough but quite exciting..real programming wud be quite tough for it, so enjoy the theoretical part for now i guess :p..
btw davis has been realized :)

Uni said...

Er, okay

You read that? Yay!

And thanks!

Theoretical part is too imaginative. I am dying to work on something 'real' :(

And yes, congrats for our nation's leaders/others finally turning into pure invertebrates. No spine left at all.

majworld said...

haha..prepare urself to work on imaginative stuff from has a big portion of it..:p
nd ya, dat was really shameful act by the govt..

Uni said...

Yea, I kinda agree to the fact that research is usually very imaginative.

Totally. :S