Partisan: Enabling Cloud-Scale Erlang Applications

arxiv.org

88 points·cmeiklejohn··7 comments

I wonder how does Partisan compare to Scalable Distributed Erlan proposed by Natalia Chechina team?

http://www.erlang-factory.com/static/upload/media/1394469405...

The problem is joining multiple clusters - you really don't want to do all P2P gossip in such cases, does Partisan address this somehow ?

Partisan builds upon HyParView for membership which seems to address many of the issues with connection scaling over fully connected meshes (section 3.3.4). It'd be interesting to see these benchmarks run with various faults injected.

I'm not sure why you'd want to avoid p2p gossip completely in these cases but they address the trade-offs by using a combination of protocols.

Direct link to the github : https://github.com/lasp-lang/partisan

The philosophy of Erlang is a single cathedral. In fact the internal part of Erlang VM itself could be viewed as a distributed system, which is designed decades ago, far more before stuff we're familiar with today.

Is it possible to limit/whitelist the gen_server’s that nodes can communicate with? Particularly for limiting code replacement / system calls. With message forwarding features it seems plausible to do. That’d be great for siloing data systems but still utilize an Erlang distribution layer, especially in systems with GPDR or HIPAA requirements. Or perhaps IoT applications.

Is this production ready? Are there any real-world examples?

AFAIK it is productiom ready and there are a couple of real use case in prod. Both with and without Lasp. Chris would know more.

Short answer yes.

Reply on news.ycombinator.com