[Gmsh] feature request: gmsh reads input from TCP port

[Christophe Geuzaine]

Al Danial wrote:
> The examples in utils/solvers/* show how one might tie an external
> solver to gmsh.  From what I can tell, this scheme means gmsh and 
> the external solver need to run on the same machine for the socket 
> communication to work.
> 
> What would be really nice is the capability to run gmsh on another
> computer on a network then send it commands over TCP.  You'd start
> gmsh on Machine_A with something like
> 
>    gmsh -p 19000
> 
> which says "wait for commands on port 19000", then from another
> computer, Machine_B, you'd connect to Machine_A's port 19000 and
> send it strings like
> 
>   "lc = 0.009;"
>   "Point(1) = {0, 0, 0, 9.e-1 * lc};"
>   "Point(2) = {.1, 0,  0, lc} ;"
>   "Point(3) = {.1, .3, 0, lc} ;"
>   "Point(4) = {0,  .3, 0, lc} ;"
>   "Line(1) = {1,2} ;"
>   "Line(2) = {3,2} ;"
>   "Line(3) = {3,4} ;"
>   "Line(4) = {4,1} ;"
> 
> (lines from t1.geo for example).  In other words, rather than reading
> input from a .geo file, in server mode gmsh would read input from a
> port.  The two machines merely need to be on the same network; don't
> need to share a file system (or operating system for that matter).
> 
> The primary benefit is that you could run gmsh on a computer that has
> optimal graphics hardware--and tied to a massive screen in a conference
> room perhaps--while running the solver on a cluster or some other
> computer with beefy CPU's but little in the way of graphics.
> 
> I can provide demo client/server TCP code if there's a possibility
> that this request might be implemented.        -- Al
> 

Hi Al - This is a really great idea! I think it should even be possible 
to implement it quite easily by using the current GmshServer.h code 
(which already allows to send/receive messages from remote hosts using 
TCP sockets).

We could for example add an option to always listen on a given port 
(instead of only listening when a user clicks "Run" in a solver dialog), 
and add new message headers to allow the transfer of actual data over 
the socket instead of just the transfer of short messages (now we only 
have "CLIENT_INFO", "CLIENT_OPTION", etc.).

Let me know what do you think.

Christophe


-- 
Christophe Geuzaine
Applied and Computational Mathematics, Caltech
geuzaine at acm.caltech.edu - http://geuz.org