MQTT Client

[cdbc]

Hi
So all this MQTT-Circus started with me *not* doing 'LNet', for networking...
In order to come full circle, I /forced/ LNet upon myself and made a plugin for LNet as well  ...I was right, /I really don't like LNet, think it's weird in a roundabout sort of way/, but hey - now it's done and I don't have to touch it again, in a foreseeable future 
As always, tested to work against Mosquitto and my own broker...
I think now is the time for me to dive deeper into the MQTT-Abyss and have a 'LookSee'...
Let me know what you think and don't hesitate to ask questions.
Regards Benny

[cdbc]

Hi
@Thaddy: Your 'supercomputer' is rambling around in my head 
- What do you use it for?
- How do you partition a collective workload?
- I guess, mqtt is used for comms amongst them?
- All the other questions, you can think of, I probably have?!?
Curious as h*ll, is what I am...
Please spill...
Regards Benny

[Thaddy]

Depends. currently it is running dummy loads to test some MPC algorithms at the protocol level. When that is stable I will try to implement some MPC tasks from the theory.
I started off with the examples from the Magpi from a few years ago. I built it just out of curiosity with no predefined plans for it. Maybe run some AI models locally.

[cdbc]

Hi Thaddy
Cool 
Quite an interesting little project...
Regards Benny

[Thaddy]

BTW my mqtt code is based on a synapse version, but I am rebasing it to use fcl-net and fcl-web

[cdbc]

Hi Thaddy
I used the /old/(er) fcl-net, because that's what's in 3.2.2, had a look at the newer 'fpsockets' by @Warfley, but I actually don't much like 'Generics'...
I don't think they should be *smeared* allover code, just because you can.
...And the generics libraries pull in a lot of code, that I mostly do not need.
I actually like fcl-net and Synapse the most 
Just my 2 cents worth.
Regards Benny

[Thaddy]

It is more that I prefer fpc-net and fcl-web since they are actively maintained.
And fcl-net has some of its roots in synapse anyway. But yes, I am mostly on trunk/main and that can differ considerably from 3.2.2/3.2 fixes.

[dbannon]

I use mqtt for controlling a cluster of rpi's ( 16 x Raspberry Zero 2 W for 64 cores in total , super computing for less than 250 euro)

In early RasPi times, I designed, on paper, a rasberrypi supercomputer as an April Fools Day joke for our newsletter. At the time, I ran the largest supercomputer south of the Tropic of Cancer (clever claim, most big SCs are well north to cheapen cooling).

My machine used tens of thousand RasPis mounted in a modified IBM tape robot, the idea being that reliability of the Pi's (OK in tens to hundreds, poor in the thousands) could be managed by automatically replacing any faulty hardware. It triggered a storm of suggestions about how we could capture a memory image before replacing a Pi so as to avoid disrupting a large MPI job. Some actually sounded (slightly) feasible.

Sadly, some people did not see the joke, triggering a series of complaints that VPAC was wasting money developing a 32bit ARM system 

Davo

[Thaddy]

Well, these clusters such as I made exist since the early Raspberry Pi one days, more than 10 years ago. So the April fool's joke did not land for many, because these clusters were not a joke at all but a genuine introduction in MPC (massive parallel compute). Several universities and scientific institutions built RPi clusters for educational purposes. (Cambridge, MIT, Delft to name a few). It is a nice toy to have. The nice thing is that because of mqtt it scales:if you have the odd $250 lying around, build the same contraption, and feel like it, we can add clusters on the go, independent of location. Note power supply is more expensive than the boards. My setup is relatively small, I know of some really silly numbers, at least 5 times mine.
Btw: it is not about the speed, it is about understanding MPC. You would not expect a toy train to reach the same speed as a TGV or Shenzen.