GDT::Grid:: Home  |  External Hyperlinks
[Yearly Archives] 2011  |  2010  |  2009  |  2008  |  2007  |  2006  |  2005  |  2004
Grid Utilitarian
"greatest happiness of the greatest number"
{created 03 October 2004}

table of contents
From IOPS To Cores To FlOPS
While working on the HPC portion of my "Learning About the Future" talk, I was interrupted with the following two news items:
   (1) Fusion-io announced they reached the one billion IOPS milestone. 
   (2) Tilera will be releasing a 100 core processor later this year.

Fusion-io employs Steve Wozniak as a Chief Scientist.

Tilera is a spinoff of MIT professor Anant Agarwa who is the director of MIT's CSAIL (Computer Science and Artificial Intelligence Laboratory).

And speaking of the HPC portion of my "Learning About the Future" talk...

HPC/21st Century Informatics

Don't ask me about the 'P' in HPC. Let 'P' equal Performance such that HPC stands for High Performance Computing. [supercomputing]

Data processing, information technology, Informatics... 21st century Informatics is HPC-based Informatics.

   Moore's Law: Number of transistors on a chip doubles every two years.

   Moore's Law: Processors get twice as fast every 18 months.

Computing roadmap: Exaflops by 02018-02020

Late last year (02011) Japan's K Computer was #1 on the TOP500.org list rated at 10.51 petaflops.

   peta-: metric prefix for 10^15
   FlOPS: Floating-point Operations Per Second
          (floating-point implies real numbers)

   10.51 petaflops = 10,510,000,000,000,000 flops
   (ten quadrillion five hundred ten trillion flops)

   10.51 petaflops Nov 02011    (10.2x in 3.5 years; 942% increase)
    1.026 petaflops Jun 02008   ( 7.5x in 3 years; 650% increase)
    0.1368 petaflops Jun 02005  (i.e. 136.8 teraflops)

My own calculations...

   petaflops     when
   ===================
     10        2011.00 <-- 10 petaflops (right now)
     20        2012.25
     40        2013.50
     80        2014.75
    160        2016.00
    320        2017.25
    640        2018.50
   1280        2019.75  <-- 1.28 exaflops

Moore's Law growth factors range from 18 to 24 months; however, my calculations show petaflops doubling every 15 months and getting us to exaflops during 4th-quarter of 02019. [Computing roadmap: Exaflops by 02018-02020]

Huge quantity (think infinite) of bits are collected by sensors/devices (cameras, scanners, medical hardware, RFIDs, nanosensors, et. al.) and are piped into supercomputers having 99.999% up-time, high-speed Internet connects (bandwidth) and huge amounts of storage (think infinite).

"We're all aware of the approximately 2 billion people now on the Internet - in every part of the plant, thanks to the explosion of mobile technology," IBM's chairman, Samuel Palmisano, said in a speech last September (02011). "But there are also upward of a trillion interconnected and intelligent objects and organisms - what some call the Internet of Things. All of this is generating vast stores of information. It is estimated that there will be 44 times much data and content coming over the next decade... reaching 35 zettabytes in 02020. A zettabyte is a 1 followed by 21 zeros. And thanks to advanced computation and analytics, we can now make sense of that data in something like real time."

It is possible that a computing cloud is a cluster of supercomputers and that the Internet morphs into a network of networked clouds.

When "stuff" can be converted into 0s and 1s, then that enables that "stuff" to be processed by HPC systems. Example: DNA converted into letters (ACAAGATGCCATTGTCC...), letters get converted into numbers (A=65, C=67, etc.) and numbers are converted into bits (binary digits).

Again, assume the 'P' in HPC is for "Performance".

   input  function  output
   =======================
   data -> HPC -> nothing
   data -> HPC -> No or Yes 
   data -> HPC -> a number
   data -> HPC -> set of numbers
   data -> HPC -> paragraph of information
   data -> HPC -> 1 page report, 2 page report, ..., 100 page report, ...
   data -> HPC -> high-performance visualization system

   data -> noise filter -> 99.999% signal 

Quoting self... What does a scientist say when you give them a petaflops supercomputer? More flops, please.

   STEAM (Science, Technology, Engineering, Arts, Mathematics)
   ===========================================================
   + more data (inputs)
   + more variables & the variables have larger domains
   + simpler algorithms because brute force becomes an option
   + processing can produce larger ranges (outputs)
   + "see" what happens when systems approach zero and infinity
   + no limit on the number "what if" scenarios
   + data sets (inputs/outputs) are archivable because of infinite storage

[23 January 2012, top]

Top 500 List For November 2011
10.51 quadrillion petaflops is #1.
   #1   Japan     10510
   #2   China      2566
   #3   U.S.       1759
   #4   China      1271
   #5   Japan      1192
   #6   U.S.       1110
   #7   U.S.       1088
   #8   U.S.       1054
   #9   France     1050
   #10  U.S.       1042
Top500.org::November 2011

[16 January 2012, top]

About the Grid Utilitarian
The Grid Utilitarian is a blog devoted to high-performance computing. This includes grid-based utility computing and 21st century Informatics. This blog was created on 3 October 2004 and it started 2012 with 243 postings.

Grid Utilitarian Archives: 2011 | 2010 | 2009 | 2008 | 2007 | 2006 | 2005 | 2004

[01 January 2012, top]


Creator: Gerald Thurman [gdt@deru.com]
Created: 03 October 2004
Last Modified: Monday, 23-Jan-2012 14:12:31 MST