- 6.1 About Ethernet
- 6.2 About Hubs, Switches, and Routers
- 6.3 About TCP/IP
- 6.4 About Packets
- 6.5 About Remote Procedure Calls (RPCs)
- 6.6 Slop
- 6.7 Observing Network Traffic
- 6.8 Sample RPC Message Definition
- 6.9 Sample Logging Design
- 6.10 Sample Client-Server System Using RPCs
- 6.11 Sample Server Program
- 6.12 Spinlocks
- 6.13 Sample Client Program
- 6.14 Measuring One Sample Client-Server RPC
- 6.15 Postprocessing RPC Logs
- 6.16 Observations
- 6.17 Summary
- Exercises
This chapter is from the book
This chapter is from the book
This chapter is from the book
6.2 About Hubs, Switches, and Routers
Point-to-point Ethernet connections between more than two machines require some form of switching fabric. There are three different kinds that you may encounter.
A hub with N links is a very cheap and now rarely used design that just reproduces one incoming transmission on all inactive outgoing links. If two or more links have incoming transmissions at once, only one is copied, and the others are dropped. Since a hub can copy only one transmission at a time, it is a shared resource like the original coax.
Switches with N links store packets at each incoming port and immediately forward them to one or more outgoing ports. Smarter switches keep tables of which MAC address destinations are attached to which port and forward only to the right destination port. A switch may store as little as two packets per incoming port, forwarding one while a second arrives. If multiple packets on different incoming ports have the same outgoing port and there is not enough buffering for all of them, some of the packets are dropped. As mentioned, Ethernet does not guarantee packet delivery, just best-effort.
A router is a more complex form of switch, using not only the MAC address in each packet but also higher-level IP and other header address information to select the output port for each packet. Routers are often connected to other routers so that a packet may go from one end node through several routers to another end node:
A ⇒ Router1 ⇒ Router2 ⇒ Router3 ⇒ B
Typical use in a datacenter is to have 40–50 servers mounted in a vertical rack with a router at the top (or middle) each rack. Traffic between servers within that rack is delivered directly from the top-of-rack router, while traffic destined for other racks is sent from the source top-of-rack router to one of several intermediate routers that eventually send the packet to the destination top-of-rack router and on to the destination server. Often in this case, the cross-router links run at a higher speed than the individual server links: for example, 10 Gb/sec copper-wire server links within a rack and 100 Gb/sec fiber-optic cross-rack links. We will use the phrase on the wire to refer to bit transmission over any kind of link. Routers often have several packets of buffering per input port, so can handle a modest amount of network congestion with several input packets destined for the same output port.
Our sample servers each have a 1 Gb/sec Ethernet port, and several are connected via a five-port switch, four ports for up to four sample servers and the fifth port connected to the rest of the building, as shown in Appendix A.