Q & A on Hardware Technologies

Q: CNAI, p.32D. Does the fact that Ethernet uses CSMA/CD preclude an Ethernet system from using frequency division multiplexing? That is, does the existence of one carrier wave on the ether keep every other host off?

A: Yes. When one host is transmitting on a segment (or a hub) other hosts refrain from transmitting. You can imagine that they all use the same "frequency". Thus, there is no FDM.

Q: Is X25net a connection oriented technology?

A: Yes. Like ATM, X.25 is connection-oriented.

Q: When you state that ' "tunneling" treats packets like data' do you mean this technique treats packets like a continuous stream of data where individual packets are sent contiguously, one after another?

A: Technically, the difference between *tunneling* and *encapsulation* is that encapsulation sends datagrams over hardware frames, while tunneling sends datagrams over some "higher level" protocol system. In terms of X.25 the term *tunneling* is appropriate because X.25 adds several layers of protocols on top of the underlying hardware.

Whether the interface is stream-oriented or message-oriented is completely independent of the underlying transmission scheme (e.g., TCP offers a stream-oriented interface, but sends data in individual segments, with one segment per datagram). Thus, one can tunnel over a stream-oriented interface or a message-oriented interface.

Q: LAN's vs. WAN's. Can the nodes of a LAN be a subset of the nodes of a WAN (or are they disjoint sets)? Are WAN's ever aggregates of LAN's?

A: The quick answer is that because LANs and WANs use completely different technologies, one cannot build a WAN by aggregating LANs. However.... some WAN technologies provide an interface that *appears* to be a LAN. For example, if one leases a satellite channel between San Francisco and new York, the equipment might appear as an Ethernet bridge to the computers using it. That is, a computer in San Francisco plugs into a 10BASE-T port, and appears to send and receive Ethernet frames. The bridge takes each transmitted frame, encodes it for transmission, sends the result across the satellite channel to New York, which then decodes the frame and delivers it to a 10BASE-T port. The advantage of using a 10BASE-T interface is that the attached computers can use standard hardware and software. That doesn't mean there's an Ethernet underneath (i.e., it's only the interface that appears to be an Ethernet). By the way, the same is true of ISDN and DSL technologies -- they accept and deliver Ethernet frames.

Q: Is it correct that a computer (host or router) has a physical address, but that its _connection_ to the Internet has an IP address? Or--is the physical address actually associated with the connection also?

A: Every network interface card has a physical address. If a computer connects to more than one network, it has more than one physical address. Thus, a router *always* has multiple physical addresses.

Q: Is the mapping from IP addresses to physical addresses one-to-one? In particular, if a host is multi-homed with two IP addresses, must it also have two physical addresses?

A: In almost all real internets the mapping is one-to-one. People sometimes play games where they assign *more* than one IP address to a given hardware interface (e.g. to simulate two subnets on the same physical wire). These many-one configurations fall outside the original intended use of IP. :-)

Q: In volume I you assert that routers must choose between routes employing physical networks that have properties of delay, throughput, and reliability. How do routers actually store this information? Is there some kind of static evaluation function the output of which characterizes the 'goodness' of a prospective route?

A: Although there has been much research on routing, the real Internet usually chooses the route that minimizes the hop count. Thus, the *type of service* routing field in the IP header is seldom used and the type of service routing you describe is seldom implemented. If it were, routers would need to keep detailed information about the throughput and delay along all paths. Delay changes so quickly that keeping accurate information is impossible.

Q: You state that each hardware technology imposes its own upper bound (MTU's) on frame size (e.g., Ethernet 1500, FDDI around 4K). Are all frames emanating from a given host 'A' of exactly the same size all the time? Do all hosts attached to Ethernet nets send frames of size 1500?

A: No, in most technologies, a frame is only as large as needed. For example, if you need to send 200 octets across an Ethernet, the outgoing frame will have a header followed by exactly 200 octets of data. For example, if a 600-octet datagram is sent over an Ethernet, the resulting frame has 600 octets of data following the Ethernet header.

As a practical matter, TCP/IP tries to optimize efficiency. Thus, if you use TCP/IP to transfer a file across an Ethernet, IP will create datagrams that are each 1500 octets long.

Another technical point: some technologies require a minimum frame size (so the hardware can distinguish real frames from a small burst of electrical noise or so the MAC protocols function properly). For example, Ethernet requires a frame to contain at least 64 octets of data. If the data being sent is less than the minimum size, it must be padded (e.g., with zeroes).

By the way, if a technology sends packets of a fixed size, they are called *cells*, not *frames*. For example, because each packet sent by ATM is the same size (53-octets), they are called *cells*, and ATM is known as a *cell-switching technology*.

Q: Please verify: when a host sends datagrams, they can be of *varying* size, but the frames emanating from the host will always be the same size.

A: Not correct -- both datagrams and frames can be of *varying* size.

Q: Are there actually two routing algorithms employed for forwarding data: one for frames on the physical net and one for datagrams on the virtual net?

A: In most LAN technologies, there is no "routing" -- one sends the packet across the LAN, all computers receive the frame, and all interface cards except the intended destination discard the frame. In most WAN technologies (at least the WANs that connect multiple hosts), the WAN needs to route frames among the switches that make up the WAN. Thus, when you send a packet across an X.25 network, switches inside the network route the packet from the port where it entered the network to the port where it exits.

So, the general answer is that routing occurs only at the IP level, not at the physical net (a handful of WAN technologies being the exception).

Q: Bridges vs. Routers: Can a bridge attach two different types of LAN's say, an Ethernet and a CDDI? How about an Ethernet and an FDDI?

A: In general, no. Bridges do not change the *shape* of the frame or the addresses. Thus, the technologies must have compatible frame formats and compatible addressing schemes. Thus, one could bridge CDDI and FDDI, but could not bridge Ethernet and FDDI.