NOTE These post series is merely a backup of some notes from my PhD sometime in 2016
Hmmm… João Barreto helped me a lot!
The idea of managing the propagation of updates between replicas depending on network conditions is quite good!
I’d see it as some sort of max-flow where everyone is a producer and also a consumer of every other node. Conditions are quite dynamic as well.
Challenge is how to interpret the problem, what are the relevant inputs, and then perform some tuning.
Some ways to alter the naive propagation methods:
- Use different paths (bmx? Bgp?…)
- Use multiple paths (partition updates and send them mptcp style)
- Aggregate and reduce at a host (MST if bandwidth is low)
- What else?
Probing for network conditions, one can use Vivaldi (thank you, Manos). For large-scale/planetary systems we can tune Vivaldi to use a lot more resources, as links are typically quite large.
Share onTwitter Facebook Google+ LinkedIn
Leave a Comment
Your email address will not be published. Required fields are marked *