[evla-sw-discuss] Delay models to station boards

[Kevin Ryan]

On May 15, 2008, at 11:57 AM, Bill Sahr wrote:
>
> Barry's worst case: For 27 antennas, 4 station boards/antenna, OTF
> mapping at a high frequency - 600 models per second.  I've seen your
> (Kevin's) email on the testing for sending UDP packets (nice job,
> interesting figures), but haven't seen anything for REST/TCP.
>
> Bruce's email speaks of a max XML message size of 14KB.  So, 600
> models/sec, @ 14KB per model (worst case)?  8,400KBytes/sec, i.e.
> 8.4 MBytes/sec ?  That's probably an overestimate.  I have little
> feel for REST/TCP capabilities with regard to data rate.  Can you
> supply some figures ?

I have no metrics but the capacity of the network will be the  
overriding factor.  TCP has the overhead of establishing the  
connection that UDP does not.  The REST portion of it I think would  
be inconsequential compared to the network and would not be too much  
more than handling data that was received via UDP.

I agree that it would be interesting to saturate a TCP connection and  
see what happens and measure the actual throughput.  Maybe after OTS  
testing, I'll do this.

What makes TCP desirable in our large EVLA/WIDAR network environment  
is taken from a book*:

TCP operates dynamically with the network:
  - Transmissions are automatically paced to the capacity
    of the network, and re-transmitted as necessary if
    not acknowledged.
  - All available bandwidth is used without saturating the
    network or being unfair to other network users.
  - TCP rapidly and reliably adjusts to changing network
    conditions - varying loads and routes.
  - The application cannot receive corrupt or out-of-sequence
    data, or data 'holes'.

*"Fundamental Networking in Java", 2006 Springer, Esmond Pitt