Graph neural networks: a review of methods and applications Zhou et al., arXiv 2019
It’s another graph neural networks survey paper today! Cue the obligatory bus joke. Clearly, this covers much of the same territory as we looked at earlier in the week, but when we’re lucky enough to get two surveys published in short succession it can add a lot to compare the two different perspectives and sense of what’s important. In particular here, Zhou et al., have a different formulation for describing the core GNN problem, and a nice approach to splitting out the various components. Rather than make this a standalone write-up, I’m going to lean heavily on the Graph neural network survey we looked at on Wednesday and try to enrich my understanding starting from there.
For this survey, the GNN problem is framed based on the formulation in the original GNN paper, ‘The graph neural network model,’ Scarselli 2009.
Associated with each node is an s-dimensional state vector.
The target of GNN is to learn a state embedding
which contains the information of the neighbourhood for each node.
Given the state embedding we can produce a node-level Continue reading
SD-WAN allowed the sensor manufacturer to streamline and automate operations, and ultimately make it in a rapidly changing industry.
Continuing Integration and Continuing Development (CI/CD), and containers are both at the heart of modern software development. CI/CD developers regularly break up applications into microservices, each running in their own container. Individual microservices can be updated independently of one another, and CI/CD developers aim to make those updates frequently.
This approach to application development has serious implications for networking.
There are a lot of things to consider when talking about the networking implications of CI/CD, containers, microservices and other modern approaches to application development. For starters, containers offer more density than virtual machines (VMs); you can stuff more containers into a given server than is possible with VMs.
Meanwhile, containers have networking requirements just like VMs do, meaning more workloads per server. This means more networking resources are required per server. More MAC addresses, IPs, DNS entries, load balancers, monitoring, intrusion detection, and so forth. Network plumbing hasn’t changed, so more workloads means more plumbing to instantiate and keep track of.
Containers can live inside a VM or on a physical server. This means that they may have different types of networking requirements than traditional VMs, (only talking to other containers within the same VM, for example) than other workloads. Continue reading
The new year is now in full swing and we’re excited about all the great content we’ve shared with you so far! In case you missed some of it, here’s our Cumulus content roundup- January edition. As always, we’ve kept busy last month with lots of great resources and news for you to read. One of the biggest things we announced was our new partnership with Nutanix but wait, there’s so much more! We’ve rounded up the rest of the right here, so settle in and stay a while!
From Cumulus Networks:
Cumulus + Nutanix = Building and Simplifying Open, Modern Data Centers at Scale: We are excited to announce that Cumulus and Nutanix are partnering to build and operate modern data centers with open networking software.
Cumulus Networks Strengthens Board of Directors Amid Record Growth and Market Adoption of its Open, Modern Networking Software: Former Deutsche Bank Group COO, Kim Hammonds, joins board as company leads the transition to open networking and data center modernization
Moving a Prototype Network to Production: With prototyping production networks, the network becomes elevated to a standard far superior to the traditional approaches.
Operations guide: We thought it would be great Continue reading