14-760: Adv. Real-World Networks Switching fabric,

Transcript:14-760: Adv. Real-World Networks Switching fabric,:
14-760: Adv. Real-World Networks Switching fabric, Routers, Routing Lecture 5 * Spring 2020 (Kesden) Remember Socrative https://api.socrative.com/rc/FX5GFK Where are we? Last class Reviewed the basics of networking. Hopefully a refresher from prior class(es) Today Examine the role of switches and routers vis-à-vis data center and Internet design Much of this will likely be a review for some (but not nearly all) Next class Look at the internals architecture of switches and routers. This is likely totally new to most. Today Definitions Data Center Traffic Patterns Data Center Fabric Managing Parallelism Within Networks Routing and Routers/Internet Switches (This is probably mostly review) Looking forward to next class Definitions for Today “Routing” (Network-layer packet switching) The hop-to-hop management of the movement of traffic among networks “Switching” (Link-layer frame switching) The movement of frames within a local network Note: There is nothing wrong with using the word “switching” at other-than the link layer, e.g. network layer switch, application layer switch, etc. Such references describe switches that make decisions based upon information present at other layers of the network stack That just doesn’t happen to be the way I am using the words today, so I define them here for today’s discussion. Today Definitions Data Center Challenges and Traffic Patterns Data Center Fabric Managing Parallelism Within Networks Routing and Routers/Internet Switches (This is probably mostly review) Looking forward to next class DC Networking Challenges Within a Data Center: Distances are short, the speed of light is less of a factor Serialization/Deserialization delay (bits/second) is a factor Managing bandwidth becomes complex as it is easy to add wire, but hard to use it. vs. Long haul: Speed of light is significant source of latency, unlikely to be resolved soon Additional bandwidth can typically readily be achieved with parallelism, e.g. more fiber Types of Traffic Ingress – Traffic entering from the outside, e.g. requests entering a data center Most commonly from outside requests. Can be from queries to related off-site services. Egress – Traffic leaving to the outside, e.g. responses, streaming video, database results, etc. Most commonly from providing service. Can be queries to related off-site services East-West – Traffic within a data center that crosses higher level switches Most commonly from communication within integrated services Incast – Traffic within a data center that becomes concentrated Most commonly from parallel query concentrating results Types Of Traffic: Ingress Core Aggregation Leaf Types Of Traffic: Egress Core Aggregation