design guidance

[ma740988]

I inquired about utilizing a vector of pairs just yesterday and after
receiving some feedback, I got to thinking my thoughts with regards to
my initial 'approach' was flawed to begin with:

Here's my 'dilema' in a nutshell. I've got 24 sources ( call them 0
... 23 )
sending data to a recipient ( call this 'X' ). The data from each
source has
what's called an 'id' (call it segment id) assosciated with it.

Today, I've got a maximum of two id's ( 0 and 1 ).
X will receive from all sources, a source id, a segment id and data
size. So now:

struct from_source_to_ x { // transmitted from all sources (
0...24) to X
unsigned int the_segment;
unsigned int the_source_id;
unsigned int the_size_of_dat a;
};

Upon receipt of the data from _all_ sources, X will perform processing
on the data. X knows (via a structure that's sent to it intially )
that the processing order is segment 0 followed by segment 1. i.e
process segment 0 first followed by segment 1.

So on X, the very first thing X receives before anything is a table
that tells X the processing order:
segment what_to_do
0 IFFT // I want to an IFFT
1 FFT // do an FFT here

The trouble on X is all the 'required house keeping that goes along
with receipt of the segments' from the 24 sources.

Having said that - on X (i.e local to X) - I need to track three
things.
1. The size of the segment per source
2. The address where I store the segment - for _each_ source on X
3. A count parameter to keep track of things.

In it's simplest form. Lets assume source 0 transmits segment 0, with
a size of 0x10000;

// so now source 0 fills out the from_source_to_ x composite
from_source_to_ x source0;
source0.the_seg ment = 0; // I source 0 just did something
with data on segment 0
source0.the_sou rce_id = 0; // Here's my source_id
source0.the_siz e_of_data = 0x100000; // Here's the results size of
what I came up with

Upon receipt of the the above header. The code executing on "X' will.
Increment the count parameter associated with _that_ source (i.e
source 0), store the size of data from source 0, and the keep track of
the offset in memory where I stored source 0's data.

Now do that X will do that the remainder. i.e receive data from a
source, increment the count, store the addresss of where the data was
moved and the size of the data.

If the count == 24, then X has received results from every source - for
segment 0.
if ( count0 = 1 && count1 = 1 && .. up to count23 == 1 )
// I've received all the data.
// do an IFFT on all the data.

X then will do an IFFT on the data from all the sources. Once 'X' is
complete with its' processing on segment 0. X will 'request' segment
1 from the sources. I'm executing in a
multi-threaded environment but enviornment 'details' aside, that's it
in a nutshell.

Now here's where it gets tricky for me - the design aspect of this.

Pictorially my thoughts were a map where first would be the
'what_to_do', second is a struct that entails - segment id, sources,
source counts, source data size, offset in memory. So now:

__ what_to_do_ segment_id
IFFT 0
0........ 24 (24) sources
0 ....... 0 (24) counts - all
zeros initially
X1.......X24 (24) data size - this
we'll get from every source
Y1......Y24 (24) address in memory
- this part is tricky but doable.
Trouble is, something just doesn't seem right about the use of map
here.

That said, I'm interested in design guidance. After think about this.
I'm thinking a chain of responsibilty pattern might be prudent here.
Bottom line: I'll take a suggestion (source snippet) and will go off
an implement it.

I tend to struggle with getting the 'design going'. The real
tricky/confusing aspect of this is trying to detemine if
1. I've received all the results.
2. How do I now, 'walk' through a struct or (?) and get the address of
every source and the size of the data. Whew!!

Thanks in advance.