Let’s start with AWS’s new networking philosophy, which focuses on making community connectivity almost invisible to customers and directors alike. For AWS, networking must be as dependable as flipping a change—it merely works, and nobody notices until it fails. To fulfill this lofty objective, AWS spent the previous decade transferring away from conventional, proprietary community {hardware} and has constructed a unified, customized stack that spans every thing from silicon to software program.
On the coronary heart of this innovation is the choice to make use of a single switching application-specific built-in circuit (ASIC)—the silicon “mind” on the middle of each community change—throughout its aggregation, core, and border networks. Reasonably than the outdated business apply of blending and matching {hardware} from varied distributors (every with its personal firmware, bugs, and scaling challenges), AWS has community engineering and operations groups concentrate on a single, constant basis. This not solely simplifies procurement and troubleshooting but additionally permits scalable, reproducible deployments that surpass the velocity of typical enterprise or cloud architectures.
All these switches run NetOS, a Linux-based working system that AWS constructed to optimize safety, automation, and speedy patching. If a bug seems, AWS can repair it in all places, immediately; there’s no ready for vendor patches and no danger that yesterday’s firmware issues may spiral into tomorrow’s outage.
The numbers are staggering. The present AWS change handles 51.2 terabits per second (Tbps) throughout 64 ports, every working at 800Gbps. However the tempo of innovation is accelerating: AWS’s next-generation change, coming quickly, will attain 102.4Tbps, with ports operating at 1.6Tbps. Mixed, AWS’s community contains roughly 2 million units, 50 million to 60 million optical hyperlinks, and greater than 20 million kilometers of fiber—sufficient to circle the moon 25 instances.
