4 ms·

I've recently noticed that there's an uptick in the number of Paxos related posts. Any particular reason why the algorithm has gained more popularity recently?

gbrown_ 6y ago
Baader-Meinhof phenomenon?
- tcgv 6y ago
  I was quietly thinking that as well these past few days. You see, my last submission on HN hit front page and was on the matter of consensus protocols. After that either my awarness on the topic increased or the frequency of postings about it did.
krick 6y ago
I don't know if there is any specific reason with Paxos, but HN front-page topics always tend to be somewhat thematically-related, so no real surprise here.
kzrdude 6y ago
Distributed consensus is quite hot - like blockchain tech. This is another aspect of that?
exacube 6y ago
One of those topics which hits the right balance of sufficiently useful, interesting and understandable without too much training.
karolist 6y ago
I think it's same reason why Designing Data Intensive Applications is in the top most popular books on safaribooks for quite some time now. Many people are grinding for job interviews and many companies now copy FAANG and have a "systems design" round, Paxos/Raft is one of the key topics there, thus it's discovered by more and more people.
akiselev 6y ago
I suspect Rust and Jepsen have a hand to play in its growing popularity: personally, I'm much more comfortable implementing or using techniques, algorithms, and patterns in Rust that I would never have risked in a higher level language, let alone in C/C++. The risk calculus just never panned out in their favor because of maintenance and availability of talent concerns. The Rust libraries for Raft and Paxos are much more approachable than my past experience with other implementations and the language itself provides a higher level of confidence. Jepsen, likewise, makes testing distributed systems an approachable problem. Case in point: I'm currently experimenting with Raft and Paxos with the intent of putting it into production for a service that streams live updates to subscribers over websockets with time sensitive business logic that must only be executed once per object on a timer. Before Rust, I would have just slung the whole problem over to whichever infrastructure team was a glutton for punishment so that they could set up Consol/kafka/debezium/whatever and set a reminder three months from now to revisit the issue.
forgotmypw17 6y ago
I think it's because distributed consensus is a pattern which is growing in use exponentially as people wake up to all the disadvantages of huge centralized services.

gbrown_ 6y ago

Baader-Meinhof phenomenon?

tcgv 6y ago
I was quietly thinking that as well these past few days. You see, my last submission on HN hit front page and was on the matter of consensus protocols. After that either my awarness on the topic increased or the frequency of postings about it did.

krick 6y ago

I don't know if there is any specific reason with Paxos, but HN front-page topics always tend to be somewhat thematically-related, so no real surprise here.

kzrdude 6y ago

Distributed consensus is quite hot - like blockchain tech. This is another aspect of that?

exacube 6y ago

One of those topics which hits the right balance of sufficiently useful, interesting and understandable without too much training.

karolist 6y ago

I think it's same reason why Designing Data Intensive Applications is in the top most popular books on safaribooks for quite some time now. Many people are grinding for job interviews and many companies now copy FAANG and have a "systems design" round, Paxos/Raft is one of the key topics there, thus it's discovered by more and more people.

akiselev 6y ago

I suspect Rust and Jepsen have a hand to play in its growing popularity: personally, I'm much more comfortable implementing or using techniques, algorithms, and patterns in Rust that I would never have risked in a higher level language, let alone in C/C++. The risk calculus just never panned out in their favor because of maintenance and availability of talent concerns. The Rust libraries for Raft and Paxos are much more approachable than my past experience with other implementations and the language itself provides a higher level of confidence. Jepsen, likewise, makes testing distributed systems an approachable problem. Case in point: I'm currently experimenting with Raft and Paxos with the intent of putting it into production for a service that streams live updates to subscribers over websockets with time sensitive business logic that must only be executed once per object on a timer. Before Rust, I would have just slung the whole problem over to whichever infrastructure team was a glutton for punishment so that they could set up Consol/kafka/debezium/whatever and set a reminder three months from now to revisit the issue.

forgotmypw17 6y ago

I think it's because distributed consensus is a pattern which is growing in use exponentially as people wake up to all the disadvantages of huge centralized services.