6014: Networked Life & Data Science, Textbook: Networked Life 20 Questions and Answers
details
– Short answer with
minimum math
– Long answer with a
more formal
treatment
– Advanced materials
for further study
Network basics
Ch. 1, 18 (Introduction, cellular, and WiFi)
生僻单词
tablet
dongle
antenna
electromagnetic
attenuation
the fourth power
tesselate
non-trivial
scramble
uplink
downlink
coding gain
leverage
the ratio of the received signal power
invert
Ch. 1 What makes CDMA work for my smartphone?
cellular network
phones are used for data applications.These data fly through a cellular network and the Internet. The cellular network in turn consists of the radio air-interface and the core network.
the fundamental concept of cellular architecture
The entire space of deployment is divided into smaller regions called cells
There is one base station (BS) in each cell, connected on the one side to switches in the core network, and on the other side the mobile stations (MS) assigned to this cell.
the deployment of base stations is based on careful radio engineering and tightly controlled by a wireless provider
Why do we divide the space into smaller regions? Because the wireless spectrum is scarce and radio signals weaken over space.
Transmitting signals over the air means emitting energy over parts of the electromagnetic spectrum. Certain regions of the spectrum are allocated by different countries to cellular communications
How can the users in the same cell share the same frequency band?
How can the users in the same cell share the same frequency band? There are two main approaches: orthogonal and non-orthogonal allocation of resources.
orthogonal allocation
In orthogonal allocation, each user is given a small band of frequency in Frequency Division Multiple Access (FDMA), or a timeslot in Time Division Multiple Access (TDMA). Each user’s allocation is distinct from the others
non-orthogonal allocation
non-orthogonal allocation, allows all users to transmit at the same time over the same frequency band, as in Code Division Multiple Access (CDMA).
But how can we tell the users apart if their signals overlap with each other?
The core idea behind the CDMA standards
The core idea behind the CDMA standards is as follows:
the transmitter multiplies the digital signals with a sequence of 1s and -1s, a sequence we call the spreading code. The receiver multiplies the received bits with the same spreading code to recover the original signals. This is straight-forward to see: 1 × 1 is
1, an −1 × −1 is also 1.
What is non-trivial is that a family of spreading codes can be designed such that only one spreading code, the original one used by the transmitter, can recover the signals. If you use any other spreading codes in this family, you will get noise-like, meaningless bits. We call this a family of orthogonal codes.
direct sequence spread spectrum
Users are still separated by orthogonalization, just along the “code
dimension” as opposed to the more intuitive “time dimension” and “frequency dimension.” This procedure is called direct sequence spread spectrum, one of the two ways to enable CDMA.
uplink, where mobiles talk to the base station
downlink, where the base station talks to the mobiles
three issues we have to address in wireless channels
Interference, together with attenuation of signal over distance and fading of signal along multiple paths, are the top three issues we have to address in wireless channels.
near-far problem
a user standing right next to the BS can easily overwhelm another user far away at the edge of the cell
the basic transmit power control algorithm
multiple MSs trying to send signals to the BS in a particular cell.
The BS can estimate the channel quality from each MS to itself, （e.g., by looking at the ratio of the received signal power to the transmitted power, the latter being pre-configured to some fixed value during the channel estimation timeslot. ）
Then, the BS inverts the channel quality and sends that value, on some feedback control channel, back to the MSs, telling them that these are the gain parameters they should use in setting their transmit powers.
This way, all the received signal strengths will be equal. This is the basic transmit power control algorithm in CDMA.
Ch. 18
Ch. 13, 14 (Packet switching and Internet)
生僻单词
tricky
the messy requirements of backward compatibility,
incremental deployability,
and economic incentives.
dedicated
unicast session
chopping up
ever-increasing
decommissioned
orthogonal
throughput
jitter
Statistical multiplexing
bursty
dial-up
optic fibers
hop-by-hop retransmission
dumb
encapsulate
prefix
traversing
egress
coarse
granularity
a corporate intranet
Ch. 13 (How does traffic get through the Internet?)
Packet switching
circuit-switched networks
We can either dedicate a fixed portion of the resources along the path to each session
Each circuit in circuit switching may occupy either a particular frequency band or a dedicated portion of timeslots.
advantages
guarantee of quality
packet-switched networks
we can (a) mix and match packets from different sessions and (b) share all the paths.
there is no dedicated circuit for each session. All sessions have their packets sharing the paths.
advantages
ease of connectivity
there is no need to search for, establish, maintain, and eventually tear down an end-to-end circuit for each session.
scalability due to efficiency
refers to the ability to take on many diverse types of sessions
There are two underlying reasons for the efficiency of packet switching, which in turn leads to high scalability. These two reasons correspond to the “many sessions share a path” feature and the “each session can use multiple paths” feature of packet switching, respectively. We call these two features statistical multiplexing and
resource pooling:
statistical multiplexing
resource pooling
disadvantages
in packet switching, a session’s traffic is (possibly) split across different paths, each of which is shared with other sessions. Packets arrive out of order and need to be re-ordered at the receiver. Links may get congested. Throughput and delay performance become uncertain. Internet researchers call this best effort service that the Internet offers, which is perhaps more accurately described as no effort to guarantee performance.
Layered architecture
Distributed hierarchy
hierarchy helps take care of the large size by “divide and conquer” in terms of the physical span
hierarchical levels
tier-1 ISP
A few very large ISPs with global footprints are called tier-1 ISP
they form a full mesh peering relationship among themselves: each tier-1 ISP has some connection with each of the other tier-1 ISPs. This full mesh network is sometimes called the Internet backbone.
tier-2 ISPs
There are many more tier-2 ISPs with regional footprints
Each tier-1 ISP is connected to some tier-2 ISPs, forming a customer-provider relationship. Each of these tier-2 ISPs provides connectivity to many tier-3 ISPs, and this hierarchy continues. The points at which any two ISPs are connected are called the Point of Presence (PoP).
ISP
An ISP of any tier could be providing Internet connectivity directly to consumers. Those ISPs that only take traffic to or from its consumers, but not any transit traffic from other ISPs, are called stub ISPs. Typically, campus, corporate, and rural residential ISPs belong to this group.
IP Routing
Addressing
Routing
Forwarding
Dynamic Host Configuration Protocol (DHCP) server
As a device, you either have a fixed, static IP address assigned to you, or you have to get one dynamically assigned to you by a controller sitting inside the operator of the local network. This controller is called the Dynamic Host Configuration Protocol (DHCP) server
BorderGateway Protocol (BGP)
BorderGateway Protocol (BGP) is the dominant protocol for inter-AS routing.
Routing Information Protocol (RIP)
Routing Information Protocol (RIP) uses the distance vector method, where each node collects information about the distances between itself and other nodes
Open Shortest Path First (OSPF)
Open Shortest Path First (OSPF) uses the linked state method,
where each node tries to construct a global view of the entire network topology.We will focus on the simpler RIP in the next section, saving OSPF for Advanced Material.
Ch. 14 (Why doesn’t the Internet collapse under congestion?)
生僻单词
vicious
Principles of distributed congestion control
TCP Tahoe
End-to-end control via negative feedback
The rate at which a sender sends packets is decided by the sender
itself. But the network provides hints through some feedback information to the senders.
Sliding-window-based control
we pipeline by providing a bigger allowance. Each sender
maintains a sliding window called the congestion window (cwnd). If the window size is 5, that means up to 5 packets can be sent before the sender has to pause, and wait for acknowledgement packets to come back from the receiver. For each new acknowledgement packet received by the sender, the window is slid one packet forward and enables the sending of a new packet, hence the name “sliding window”.
Additive increase and multiplicative decrease
Infer congestion by packet loss or delay
The early versions of TCP congestion control made an important assumption: if there is a packet loss, there is congestion. This sounds reasonable enough, but sometimes packet loss is caused by a bad channel, like in wireless links, rather than congestion.
In addition, often it is a little too late to react to congestion by the time
packets are already getting dropped. The first problem has been tackled by many proposals of TCP for wireless. The second problem is largely solved by using packet delay as the congestion feedback signal. Instead of a binary definition of congestion or no congestion, delay value implies the degree of congestion.