Last modified 10 years ago Last modified on 12/13/2008 05:14:41 PM

Amdahl's Laws and LSST


Amdahl's Laws:

  1. Amdahl’s parallelism law: If a computation has a serial part S and a parallel component P, then the maximum speedup is S/(S+P).
  2. Amdahl’s balanced system law: A system needs a bit of IO per second per instruction per second.
  3. Amdahl’s memory law: alpha=1: that is the MB/MIPS ratio, in a balanced system is 1.
  4. Amdahl’s IO law: Programs do one IO per 50,000 instructions

Amdahl’s balanced system law and LSST

  • The amdahl number for supercomputers running high-performance-computing applications ranges around 1E-05.
  • Computation-heavy data-intensive simulations tend to use hardware with amdahl number around 1E-03.
  • Data intensive analytical applications are in the range of 1E-01 or even 1.

The above claims are based on Alex's talk and eBay's blog.

For LSST, based on the baseline documents outlining planned hardware purchases, the amdahl number falls in the range E-05, see the attached spreadsheet. Are we underestimating number of spindles needed, overestmating CPU needs or both?

Amdahl's memory law

[to be calculated]