Design details Was:Re: [evla-sw-discuss] RTOSes ?

[Kevin Ryan]

> Indeed.  I would think that the communications protocol will take most
> of the cycles of the MIB if you use a real TCP/IP stack. 
> 
> I have so many more questions that I will go to the Web documents and
> read them for answers, but here are a few things I first thought of:
> 
> Why Ethernet?
> 

We want to keep everything (including the interfaces) as generic
as possible.  Ethernet and CAN were the two main contenders and 
we settled on Ethernet because of the wide selection of protocols
available, its adequate speed, price and its availability in the
future.

> How are you planning to connect up all 50-odd of these MIBs in each
> antenna into a network without RFI 
>

Fiber Optic cabling everywhere.  We have not decided on the intra-
antenna (inside the antenna) distribution yet (i.e. simple hubs
vs. managed switches, etc.).

> and without breaking the bank?
> 

We are breaking the bank.  I think we are talking about $8k - $15k
per antenna just for switches.

> Isn't deterministic performance needed for the MIB->control computer link?
> 

It depends how we implement things.  If we send 'on-time' commands then
yes, it will have to be deterministic.  If we queue commands in the MIBs
then no. 

I prefer keeping away from command queues (at the MIB level) and
believe we can make it work by using a 10 ms (or thereabouts) system wide
interrupt to synchronize the 'next' commands.

> It seems the system is architected as one computer to control the whole
> array, instead of a computer per antenna.  This will make your life
> much more difficult, I think.  

I agree.  I am still pushing for a distributed system with an Antenna
Control Computer in each antenna.

> Why was the distributed approach thrown out?

Good question.

I think it started out by some people's (justified) fears of having 
a computer in an Antenna.  Even though we later realized that we would 
probably have on the order of 50 of them in each antenna, the idea of 
the benefits of housing one or a few Antenna Controllers in the climate 
friendly Control Building stuck.

It then progressed to 'Since all the antenna computers will be
located in the same physical area, there is really no need for a
separate one for each antenna ... '

Centralizing the processor(s) will force us to use managed switches
in the antennas which are noisy computers anyway. 

Also, centralizing will make running an array of different antenna
types more difficult.

This subject is still being debated.