Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > comp.programming > #3911

A scalability example

From aminer <aminer@toto.net>
Newsgroups comp.programming.threads, comp.programming
Subject A scalability example
Date 2013-10-10 17:04 -0700
Organization albasani.net
Message-ID <l374pe$53h$1@news.albasani.net> (permalink)

Cross-posted to 2 groups.

Show all headers | View raw


Hello again,

We have to become smarter by also leaning more.. and we have to apply my 
ideas, so follow with me please, what i have just say its that the 
Amdahl's law can not only allow us to calculate the optimum performance 
but also to predict the worst case scalability, so i will give you an 
example of that... so go to my following website:

http://pages.videotron.com/aminer/

And download my Parallel Archiver 1.91, and open with an editor
the file parallelzma.pas inside the zipfile, you will read the
the procedure "procedure TCallbacks.lzmaCompress1(obj: Pointer);"
you will notice that this procedure is also using a Ticket mechanism
but this Ticket mechanism do not use a proportional backoff, so
i think i must add a proportional backoff  to lower the cache coherence 
traffic, so look at the  "procedure TCallbacks.lzmaCompress1(obj: 
Pointer);" you will notice that it's easy to predict the worst case 
scalability
of this procedure, just calculate the parallel part and its maximum 
bound to compress the following stream: "Tjob(obj).stream" and calculate
the serial time to write the compressed stream to the hardisk and
after that just plug the two numbers in the Amdahl's equation, so plug 
the percentage of serial time and the percentage of the parallel time 
into the Amdahl's law and it will give you the worst case scalability, 
so as in realtime systems
this worst case scalability is important, it's the lower bound of the
scalability/performance that we can get. Other than that you have to 
know that
the serial time of the serial part can be lowered by using a faster 
SSD hardisk, so that it will allow parallel archiver to scale much more.


So as you have noticed the purpose of the Amdahl's law is not only
to calculate the optimum performance but also to predict the worst
case scalability that is also very important to know.



Thank you,
Amine Moulay Ramdane.



Back to comp.programming | Previous | Next | Find similar | Unroll thread


Thread

A scalability  example aminer <aminer@toto.net> - 2013-10-10 17:04 -0700

csiph-web