The Glade 4.0

"Turn the lights down, the party just got wilder."
It is currently Sun Nov 24, 2024 3:36 pm

View unanswered posts | View active topics


Board index » The Glade » Tech

All times are UTC - 6 hours [ DST ]



Post new topic Reply to topic  Page 2 of 3
  [ 54 posts ]  Go to page Previous  1, 2, 3  Next
  Print view Previous topic | Next topic 
Author Message
Taskiss
 Post subject:
PostPosted: Wed Dec 01, 2010 2:33 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
Yup. That's what started the whole multi-core thing in the first place.

Now all we have to worry about is Amdahl's law.

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
PostPosted: Wed Dec 01, 2010 2:43 pm 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
No, we really don't.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Taskiss
 Post subject:
PostPosted: Wed Dec 01, 2010 2:44 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
Well, you don't, but I do.

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
 Post subject: Re:
PostPosted: Wed Dec 01, 2010 2:52 pm 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
Taskiss wrote:
Well, you don't, but I do.
You really don't. If something can't be parallelized, then there's no amount of parallelization that can improve the efficiency of that operation. You're not constrained by Amdahl's Argument in any way, because it's not a problem to be solved.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Taskiss
 Post subject:
PostPosted: Wed Dec 01, 2010 3:08 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
I do systems integration - I have to deal with moving programs to new hardware all the time. More != faster, not at all.

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
PostPosted: Wed Dec 01, 2010 3:11 pm 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
It doesn't matter if you have to deal with migrating software onto new and different processing hardware. The procedural limitations of the processing threads will remain in place regardless of the parallelizing opportunities available. You don't have to worry about Amdahl's Argument.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Taskiss
 Post subject:
PostPosted: Wed Dec 01, 2010 3:14 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
Gosh, Khross! You just made my job 10X easier!

Oh, wait... no, you didn't.

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
PostPosted: Wed Dec 01, 2010 3:18 pm 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
Invoking Amdahl's Argument when the limitations you face have nothing to do with doesn't make your job easier, either.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Taskiss
 Post subject:
PostPosted: Wed Dec 01, 2010 3:22 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
Ignorance doesn't slow you down a bit. That's impressive.

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
PostPosted: Wed Dec 01, 2010 3:46 pm 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
Taskiss:

I'm not ignorant of Amdahl's Law. It simply isn't going to make your life easier. A computer's "speed" is limited by the most time consuming single threaded process in the software its running. And, assuming that process is cannot be made into multiple parallel processes, no amount of additional cores will break that ceiling. So, please, tell me how I'm ignorant of the way this works. Amdahl's Law isn't a problem to be solved.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Taskiss
 Post subject:
PostPosted: Wed Dec 01, 2010 3:58 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
I wasn't talking about ignorance of Amdahl's law, the ignorance being displayed was assumptions about the requirements of my job.

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
PostPosted: Wed Dec 01, 2010 4:04 pm 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
Taskiss:

It really doesn't matter what the requirements of your job happen to be in this case. Amdahl's Argument isn't a problem you can solve. And if you're migrating a whole bunch of single-threaded software you cannot rewrite onto multi-core, multi-processor systems, then you're limited by the new processor's single-threaded efficiency relative to that piece of software. So at what point can it make your life or job easier?

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Taskiss
 Post subject:
PostPosted: Wed Dec 01, 2010 6:23 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
I migrate programs Khross. Not applications. Programs that have multiple databases, tb's of storage, multiple net and hba interfaces... weblogic, datastage, peoplesoft components, etc. I did the calculations to size partitions when moving these programs from pa-risc superdomes to itanium superdome 2's ... up to 128 cpu's, 128 gb ram, etc, etc.

I guess I could just wing it... but I find it better to do some calculations first. And, I've got a reputation for hitting my numbers. If I tell you that you're going to have a 25% increase in performance out of your month end closing jobs after we move your program, you'll have a 25% increase.

But hey, you keep arguing with someone that's achieved success at what he does.

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
PostPosted: Wed Dec 01, 2010 7:31 pm 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
Taskiss:

Right, so you're proving your point how exactly? You're using Amdahl's algorithm for calculating speed increases in a situation where the number of sequential operations is likely to be essentially 0 or as close to 0 as possible based on the work of mathematicians and programming trends that came later. So, either way, you're not making your life any easier by invoking Amdahl's Argument and misapplying it in an internet forum.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Taskiss
 Post subject:
PostPosted: Wed Dec 01, 2010 9:27 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
I'm thinking you're out of your element, Khross. I've never seen a system with 0% sequential portions. Ever.

Here's a real world example from a Solaris internals site:

http://www.solarisinternals.com/wiki/in ... processing

Quote:
The figure below shows the results for a series of scaling benchmarks that were performed using database workloads on a large SMP configuration. These application benchmarks were performed on a single system image, by measuring throughput as the number of processors was increased.

Image

And that's just the cpu component. Amdahl needs to be used on the network traffic and the disk traffic too, when those components change.

You can't just double the processors and claim double the performance increase. Or, you can, but you'd be wrong.

Read this book then get back to me:
http://www.amazon.com/Essentials-Comput ... 076370444X
and this:
http://info.ee.surrey.ac.uk/Teaching/Co ... arch8.html
Code:
Amdahl's Law
When adding enhancements to our machine we need to know what effect this will have on performance. This is neatly encapsulated by Amdahl's law:

                  Execution time for entire task without enhancement
Speedup = ------------------------------------------------------------------
          Execution time for entire task using enhancement whenever possible

When we consider an enhancement we need to establish what fraction of the computation time can be converted to exploit the proposed enhancement. This is f_enh. We also need to consider the speedup for the task as a result of adding the enhancement. This is DT_new. The new execution time for the task is then:

T_new = T_old * (1 - f_enh) + f_enh
                              -----
                              DT_new

or in terms of the overall speedup:

T_old             1
----- =  -------------------
T_new                  f_enh
         (1 - f_enh) + ------
                       DT_new

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
PostPosted: Thu Dec 02, 2010 8:12 am 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
Seems to me that you're reading sources which don't understand what Amdahl's Argument says. It also seems to me that you're unfamiliar with various criticisms of it for large systems. Curious thing, by the way, that the Oracle DSS 4-Way Join has approximately 98% of perfect scaling, but you state there are no systems as close to non-zero sequential processes possible. You might not want to disprove your own position with your sources in the future.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Numbuk
PostPosted: Thu Dec 02, 2010 11:13 am 
Offline
User avatar

Joined: Sat Sep 05, 2009 2:40 am
Posts: 3188
Image





JK. ;-) Just trying to lighten things up.

_________________
Les Zombis et les Loups-Garous!
Top
 Profile  
Reply with quote  
Taskiss
PostPosted: Thu Dec 02, 2010 6:05 pm 
Offline
User avatar

Joined: Fri Feb 05, 2010 11:59 am
Posts: 3879
Location: 63368
Khross wrote:
Seems to me that you're reading sources which don't understand what Amdahl's Argument says. It also seems to me that you're unfamiliar with various criticisms of it for large systems. Curious thing, by the way, that the Oracle DSS 4-Way Join has approximately 98% of perfect scaling, but you state there are no systems as close to non-zero sequential processes possible. You might not want to disprove your own position with your sources in the future.

I'm not disproving anything. Those were performance benchmarks. In the real world you see quite a different picture.

Amdahl was a conservative.

I was looking for this last night and couldn't find it... but then I did today.

I've used these equations a lot in the last few years. Not so much at the new job, but that's a good thing. Gathering the info to reach a number I was confident about is a pain in the ***.

Attachment:
image1.jpg
image1.jpg [ 42.69 KiB | Viewed 1876 times ]

Attachment:
image2.jpg
image2.jpg [ 53.02 KiB | Viewed 1876 times ]

Attachment:
image3.jpg
image3.jpg [ 61.63 KiB | Viewed 1876 times ]

_________________
In time, this too shall pass.
Top
 Profile  
Reply with quote  
Khross
PostPosted: Fri Dec 03, 2010 8:17 am 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
Again, that's a misstating the purpose of Amdahl's Argument, Taskiss. You need better sources which actually understand the theoretical construct they're using, instead of applying a label that doesn't actually fit. Or, they could just state they're using Amdahl's Speedup Algorithm to project improvements in performance without considering the purpose for which it was created. Square pegs; round holes and all that.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Lex Luthor
 Post subject:
PostPosted: Tue May 24, 2011 2:12 pm 
Offline

Joined: Thu Sep 03, 2009 10:03 am
Posts: 4922
I liked this article... IBM predicts the end at 50nm.

http://domino.watson.ibm.com/comm/wwwr_ ... os398.html
Top
 Profile  
Reply with quote  
Talya
 Post subject: Re:
PostPosted: Tue May 24, 2011 5:15 pm 
Offline
Oberon's Playground
User avatar

Joined: Thu Sep 03, 2009 9:11 am
Posts: 9449
Location: Your Dreams
Lex Luthor wrote:
I liked this article... IBM predicts the end at 50nm.

http://domino.watson.ibm.com/comm/wwwr_ ... os398.html


How old is that article? 50nm thought to be the minimum. Hah...nice...considering we're now at 32nm and continuing to drop. On the other hand, they grossly overestimated clock speeds (10 GHz at 50nm?) and the difficulty to implement parallel processing.

_________________
Well Ali Baba had them forty thieves, Scheherezade had a thousand tales
But master you in luck 'cause up your sleeves you got a brand of magic never fails...
...Mister Aladdin, sir, What will your pleasure be?
Let me take your order, Jot it down -You ain't never had a friend like me
█ ♣ █
Top
 Profile  
Reply with quote  
Lenas
 Post subject:
PostPosted: Tue May 24, 2011 5:40 pm 
Offline
Web Ninja
User avatar

Joined: Wed Sep 02, 2009 8:32 pm
Posts: 8248
Location: The Tunt Mansion
August 8th, 2001. Nice and relevant.
Top
 Profile  
Reply with quote  
Lex Luthor
 Post subject: Re:
PostPosted: Tue May 24, 2011 6:34 pm 
Offline

Joined: Thu Sep 03, 2009 10:03 am
Posts: 4922
Lenas wrote:
August 8th, 2001. Nice and relevant.


Actually it's from 1998.

"That should happen around eight years from now, in 2006"

"Early in the next century, the rapid shrinking of transistors -- which has led to a doubling of computer performance every three years -- will come to an end. What then?"

I just think it's funny how even the most knowledgeable people can't predict a decade in advance about technology, even on a basic level of whether a technology will continue to improve or stagnate. Also if you read old Internet writings, it's like nobody expected high-bandwidth wireless technology to rapidly spread across the world.

It's also amusing how in this article the author wonders if there won't be demand for higher performance computers. In fact, the only utility he can think of is voice recognition, which incidentally my smart phone can handle perfectly well.
Top
 Profile  
Reply with quote  
Lex Luthor
 Post subject:
PostPosted: Wed May 25, 2011 12:09 am 
Offline

Joined: Thu Sep 03, 2009 10:03 am
Posts: 4922
Intel's roadmap. I'm really excited for 14nm!! Now I have a rough estimate of when to upgrade my computer.

http://www.anandtech.com/show/4345/inte ... om-in-2014

(large roadmap image in spoiler)

Spoiler:
Image


Devastation brought by our failing economy on business (making fun of the tin-foilers here is fun):

Spoiler:
Image
Top
 Profile  
Reply with quote  
Khross
PostPosted: Wed May 25, 2011 9:02 am 
Offline
Evil Bastard™
User avatar

Joined: Thu Sep 03, 2009 9:07 am
Posts: 7542
Location: Doomstadt, Latveria
You really don't understand economics, Lex.

_________________
Corolinth wrote:
Facism is not a school of thought, it is a racial slur.
Top
 Profile  
Reply with quote  
Display posts from previous:  Sort by  
Post new topic Reply to topic  Page 2 of 3
  [ 54 posts ]  Go to page Previous  1, 2, 3  Next

Board index » The Glade » Tech

All times are UTC - 6 hours [ DST ]

Who is online

Users browsing this forum: No registered users and 143 guests

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB® Forum Software © phpBB Group