Re: Optimization

BemusedbyQM wrote:
> "Jon Harrop" <usenet@xxxxxxxxxxxxxx> wrote in message
> news:4353e71d$0$15057$ed2619ec@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx
>> The most efficient algorithm for any given task evolves with computer
>> design. A pair of algorithms are likely to have very different
>> performance characteristics when run on a 20 year old and a new computer.
>> Which one is faster may well be different on modern technology.
>
> thats true, i'd not thought of that

The "memory gap" is one of the most important differences. When I was
learning to program (1980s), the relative costs of arithmetic operations
was as important as the time taken to load a CPU instruction or datum. Many
old textbooks contain estimates of the relative costs of + and /.

Now, memory access can be 100x slower than arithmetic and the relative cost
of arithmetic operations is almost irrelevant. This has affected the
constant prefactor in the performance of most algorithms. Some are now much
slower and others are now much faster.

For example, reference counting was originally a fast form of garbage
collection. Now, it is one of the slowest, primarily because incrementing
or decrementing the reference counts is very expensive (they are scattered
all over memory).

--
Dr Jon D Harrop, Flying Frog Consultancy
http://www.ffconsultancy.com
.

Relevant Pages

  • Re: New 2 Cryptography
    ... Newsgroups: sci.crypt ... > cryptography and I thought I'd post here to ask a few questions. ... Also there is a question of licensing one or more algorithms, ... rounds of arithmetic operations, and you wind up with something that would ...
    (sci.crypt)
  • Re: A factoring algorithm
    ... >>than the current best algorithms at the time. ... For example, Lenstra once ... >>of arithmetic operations ... Involves factorials of numbers or something IIRC. ...
    (sci.crypt)
  • Re: A factoring algorithm
    ... >than the current best algorithms at the time. ... For example, Lenstra once told ... >of arithmetic operations ... and consequently the cost, measured in bit operations, is exponential. ...
    (sci.crypt)