Great, if you end up building one, keep us updated (here or on Twitter).
1) Yes, this is DCS (Data Collection System) data. It's a very low bandwidth signal (300 or 1200 baud) that ground stations uplink to GOES on UHF. I found there are about 20k separate transmitters in some asset database I once found somewhere. Every one (or the set that's still active) transmits a tiny packet every N minutes (some every minute, others every 30 minutes). This data is also broadcast as part of the LRIT (GOES-15) and HRIT (GOES-16 and GOES-17) streams. There is no processing for this data in the goestools package yet, but you can get the originals and walk over the data. When I last looked at it, the format looked different per receiver, it is possible this is a per-vendor type of thing. Anyway, yes, you can get a copy of this data with the receiver described in the OP guide. See https://github.com/pietern/goestools/issues/11 for some more links.
2) AFAICT MODIS is a product derived from the Aqua and Terra satellites. Both are polar orbiting sats. You can look them up here https://www.wmo-sat.info/oscar/satellites/view/81 and it looks like they don't pack a "direct readout" transmitter, so you wouldn't be able to receive anything directly yourself. Compare this to the page for GOES-16 at https://www.wmo-sat.info/oscar/satellites/view/152 and you see there are a bunch of frequencies with details listed.
Depends on the mode the satellite is operating in. Typically: a full disk image every 15 minutes, the continental USA every 5 minutes, and two selected "mesoscale" regions (they will move around based on interesting events) every minute or even 30 seconds. Not all these products are relayed through this downlink though. On HRIT you'll find a full disk image across 7 different spectral bands every 30 minutes, and mesoscale images across 3 different spectral bands every 15 minutes.
Next to this data, it also contains relayed images from GOES-15 (GOES-West), Himawari-8, EMWIN data, and NWS forecasts and plots.
It's not. The SUVI data is not broadcast on the HRIT feed, only on the GRB (GOES ReBroadcast) feed. Receiving GRB requires more effort: minimum 10ft dish, dual circular polarized feed, DVB-S2 demodulator, and different software. Never tried this, but would make for a fun project.
Re: your second point, it is strictly synchronous, though since there are 8 GPUs per process (thus have 1 process per machine) the gradient reduction is done in 3 phases. First they are reduced within the process, then across processes/machines, and then broadcast within the processes.
And there's even a tiny Redis dependency (optional though) in the code to generate these results. In particular the collective communication library needs a rendezvous phase where all nodes connect to their peers. Using Redis for this is one of the options. See: https://github.com/facebookincubator/gloo/tree/master/gloo/r...