[placement][ptg] Envisioning other technologies
From the etherpad [1]:
("My" in here is edleafe) * The more we get into the complexities of modeling nested providers, the more it seems to me that we are not using the best technology for this task. My brief experiments with graph databases last year showed a great deal of promise, but the whole extraction process pushed that to the side. * Perhaps we might try some parallel development, where resources are modeled in both relational and graph DBs. Then we might be able to compare the different solutions to complex queries on the same data sets. This is an invitation to discuss other potentially useful technologies, the relative cost of exploring them, and benefits we might get from that exploration (even if we don't end up using the other tech). And anything else anyone else wants to add. [1] https://etherpad.openstack.org/p/placement-ptg-train -- Chris Dent ٩◔̯◔۶ https://anticdent.org/ freenode: cdent tw: @anticdent
On Apr 9, 2019, at 7:04 AM, Chris Dent <cdent+os@anticdent.org> wrote:
* Perhaps we might try some parallel development, where resources are modeled in both relational and graph DBs. Then we might be able to compare the different solutions to complex queries on the same data sets.
Listening today to the problems Nova will be facing getting nested providers, NUMA affinity, shared providers, RP distance, and a few others, it was sad because those issues have already been solved using a graph database instead of a relational database. So rather than continue to preach into the void, I’d like to propose a challenge. Come up with a scenario involving any of the above concerns that is currently not doable in Nova/Placement, and I will come up with a demo that shows how it can be solved using Neo4j and py2neo. My goal isn’t to show off some cool technology; my goal is to convert those of you who are skeptics about anything that isn’t MySQL into believers. So I need an example case for the demo that if you saw it working in action, would convince you to at least take the idea of a graph database solution for placement seriously. And pointing out potential hurdles, such as Neo4j on OpenJDK, or data migrations, or anything else really isn’t helpful right now. I just want to knock your socks off. In case you missed this last year, here’s some background. Link to my original demo: https://blog.leafe.com/placement-graph-examples/ Follow-up post to Jay Pipes’s questions: https://blog.leafe.com/graph-database-follow-up/ -- Ed Leafe
participants (2)
-
Chris Dent
-
Ed Leafe