December 2018 Summaries

The blog post discusses the challenges and potential solutions for running a distributed system across multiple Kubernetes clusters, emphasizing that while Kubernetes has improved in features like dynamic volume provisioning and multi-zone clusters, it still struggles with multi-region and multi-cluster setups due to networking issues. The author explores various strategies, including static pod placement, using external load balancers, direct pod IP communication, DNS chaining, and Istio, each with its pros and cons related to performance, security, and complexity. The core issue lies in establishing reliable pod-to-pod communication and stable network addresses across clusters, with different networking solutions offering varying degrees of support and feasibility. Recent advancements in projects like Istio, Cilium, and Crossplane show promise, but a universally reliable solution remains elusive, requiring time and effort for configuration and maintenance.

The 2018 Cloud Report, which investigates the performance of Amazon Web Services (AWS) and Google Cloud Platform (GCP), reveals that AWS substantially outperforms GCP across several criteria, including throughput, CPU, network, I/O performance, and cost, with AWS providing 40% greater throughput in a TPC-C benchmark test. This report, built on extensive testing involving simulated customer workloads and micro-benchmarks, highlights AWS's superior capabilities in network throughput and latency, as well as write and read performance under various configurations. Despite the high-performing results of AWS, the report acknowledges that it did not include Microsoft Azure due to resource limitations, but plans to incorporate it in future analyses. The findings suggest AWS provides better performance at a lower cost, though the report maintains a cloud-neutral stance and plans to continue evaluating multiple cloud providers as the infrastructure landscape evolves.