Attendees:
Petar Israel Albert (online) Jens
Agenda:
What we have been working on:
- Using network calculus to create arrival, departure, service curves and backlog curves.
- Tried to minimize the probability of overflow in the buffer.
- Backlog length
- Have this working
- Time in the backlog
- Implementation issues
- The last two are not probabilities
Our problems right now
- Variance
- The variance of the probabilities is basically very low, and therefore it does not matter what satellite we choose as they are are all either 0% chance or 100% chance.
- Maybe use probabilistic arrivals (H265)
- Using distributions we have to have some sort of demands to optimize for
- Joint distributions
- Both optimizing for a probability of spending less than x time in the buffer and not going over the buffer (Maybe also something with minimizing the addition load we put on the system(Choosing a satellite which is not very congested))
- Is this a thing and does it make sense?
Monte Carlo is that a good idea? Meetings more often/ regularly?
Notes:
Make a rate/latency curve THis is the important part
Make burstiness Fix arrivalrate change burstiness Fix burstiness change arrivalrate