CS73N Note on Internet Technologies
Created 2004, Material extracted by Gio Wiederhold 7 April 2006, updated 11, 13 April 2007. TLD removed 13 May 2006.
Internet Technology
How does a message travel from your dorm room to, say, your friend at MIT?
- Software in your computer decomposes the message into small packets, each with a destination and a sequence number
- It is placed on the Ethernet cable in your dorm, either by your computer or a shared server. If a collision on the Ethernet is detected, your computer and the conflict creator back of and try again later.
- A router on your Ethernet cable checks if the recipient is on your cable, if not, it forwards it to next level network.
- This process repeats until it gets to an Internet Service Provider , in your case the University. The ISPs purchase bandwidth from a commercial company. There are often intermediaries, they and major consumers will buy long-distance bandwidth from one or several of the few major backbone providers: Sprint, MCI, ...
- Router nodes at many levels of the network keep track of recent destinations, but cannot hold all possible destinations in their memories. If a destination is unknown to a router it sends the unknown address to a DNS node which keeps track of a larger number of destination, and receives the address for the best successor node towards MIT. A DNS node itself can communicate with other DNS nodes to locate unknown addresses. If intermediate router nodes fail, then the DNS will assign a new path for the packets.
- At the destination the packets are reassembled, even if they are received out of order, and the recipients mail program is signaled that a message can now be fetched.
Arpanet growth
The ARPAnet was developed starting in 1966 to provide disaster-resilient comminications for the U.S. military and its suppliers, based on concepts expressed in the early 1960ties. Support was provided by DoD's Advanced Research Projects Agency (ARPA), which had its start in 1958 as a reaction to the USSR launch of Sputnik and the threats that were perceived at the time. Many alternate plans and technologies were discussed before actual construction began. An important decision was to use small (16K) computers as node interfaces, to avoid the load and unreliability that would ensue by using the major, large computers (up to ~256K) at the centers to be connected. Actual implementation took about a year, and by December 1969 four-nodes were connected: UCLA, Stanford Research Institute (now SRI International), UCB, and the University of Utah. Each node had two connections for communication reliability.
By 1977 more than 50 nodes were connected, as shown in the ARPAnet map of that period. The network grew steadily. having access to the Arpanet gave researchers with military contracts advantages over other researchers, and competing networks were established. There was a UUNET for Unix-based research groups, a NASA net, an IBM net, and eventually an NSF-sponsored CSnet in the 1980ties.
By 1984 there were 373 hosts computers at about 200 sites on the Arpanet.. It was still possible to print a paper directory listing every one of the 14 000 individuals who had access to the ArpaNet.
In 1986 an improved NSFnet, using ARPA net protocols, linked to the ARPAnet, creating a basis for the Internet - network of networks. By 1988 a backbone connecting 13 regional networks and 170 academic institutions was operational. They used differing protocols and connecting among them was problematic. Eventually NSF took over the research support functions of the ARPAnet.
In 1995, the NSF net was transitioned to a fully commercial base. Rapid growth ensued, and usage outside of the United States new exceeds U.S. use. Now researchers use University and Research grant funds to participate. Content services are provided either for payment, by Foundations such asWikipedia, or supported by advertising revenue. By 2005 there were over 200 000 hosts on the Internet, a connection graph is displayed at the entrance to the B-Wing on the 4th floor of the Stanford Gates CS building.
Arpanet Scaling
To accomodate growth scalability was required as the network grew to millions versus thousands of nodes. Important improvements over the Arpanet were
a. three level naming: gio@earth became gio@earth.stanford.edu
b. mapping from names to IP addresses, Ipv.4as : 171.64.64.64. – 4 x 16 bits (what is yours? Click on http://www.ipv6.org/ ) performed by Domain Name Servers (DNSs)
c. caching – a big short term memory in the nodes (more on that in webpage creation)
d. next: running out of addresses if all gadgets want their own IP address: moving to http://www.ipv6.org/ IPv6 uses 128 bits, allowing 2^128 addresses: This works out to be: 340,282,366,920,938,463,463,374,607,431,768,211,456
The associated Transmission Control Protocol: TCP, specifies:
a. Packets with headers: from, to, number
b. Nodes with forwarding information tables: Because address tables got too large to keep at every node, and keep then up-to-date- they are no cached at designated nodes: Domain Name Servers (DNS). If addresses are unrecognized locally, the DNS is queried, and DNS servers can repeat tha process if they don't have that address in their tables.
Internet Service Providers (ISPs) and how to get there and to your destination
Review: To get to the Internet you typically
a) Access a local server via a Local Area Network (LAN) using the Ethernet protocol (more on that)
b) Routers connect LANs and Wide Area Network (WAN) access points.
c) Stanford buys – from your fees – bandwidth to go to WANs as needed from an Internet Service Provider (ISP)
d) ISPs ( there may be multiple layers) go to a regional network (BART here, Los Nettos in L.A.) for out region connections that cannot be handled by them, and pay them for bandwidth.
e) Natural growing improves economics of ISPs – they can serve non-geographic regions
f) The regional system networks buy bandwidth from long-distance backbone providers to get your stuff cross-country or to international destinations. They are geographic. Why?
g) Major backbone or Tier 1 providers include MCI, SPRINT, Nexttel Verizon - formerly MCI (was Worldcom from 1998 to 2003, it incorporated UUnet), AT&T, Reliance, Qwest (formerly US West), HE-NET (a Silicone valley startup focusing on fiber), Level 3 Communications, with transatlantic cables and a large European network. Worldwide backbone providers are British Telecom, France Télélcom, VSNL (creasted by privatizing Indian Radio and Cable Communications), Teleglobe (Canada - with transatlantic cables, owned now by VSNL), NTT in Japan with transpacific cables, acquired Verio. There have been many mergers and rearrngements, and this list cannot be kept up-to-date.
h) Backbone providers exchange their packets at Internet Connection Points (IXPs), as PAIX in Palo Alto, MAE-East near DC. A major IXP is in Amsterdam, Holland, share and trade bandwidth resources by informal balancing of contributed capacity versus load experienced. They share and trade bandwidth resources by informal balancing of contributed capacity versus load experienced, called peering.
i) And now back in reverse, to your correspondents computer, or to your parents’ hypermodern refrigerator? Do you know what their regional net is?
By 2004 there was much underused backbone capacity – “black fiber” some ready to be used, as fiber in the ground. Companies as Google have invested in buying that capacity, as investment as the Internet keeps growing.
Ethernet
To get to the Internet, one typically starts at a local computer, hooked up to a Local access Network (LAN).
The most common LAN technology is the Ethernet, developed at Xerox' Palo Alto Research Center, now just Parc (on the Stanford Campus) in the 1980s. Rather than having dedicated high capacity wires from each terminal to some central computer, which then must sort out who gets to transmit what to whom and when, the Ethernet is a cooperative protocol on each participating computer. The Ethernet uses a single the wire or optical fiber cable. But such a single connection can only handle one transmission (speaker) at a time unless costly multi-frequency transmission paths and interfaces are added.
The concept is similar to one seen years ago on rural telephone party lines: pick the phone up, and if you hear someone using it, hang up and wait. All computers on one ethernet share a single high capacity wire, which may be about 1000 feet long and have a hundreds of stations, i.e., computers, on it. Any computer can put a message packet at any time on that wire, and it can be read by all participants, including the routing computer, who may forward it, likely via intermediate networks, to the Internet. That works most of the time, except when two computers start sending stuff at approximate the same time. First of all, under the Ethernet protocol – like in human conversation – each participant is required also to listen and not start sending/talking when there is already someone sending/talking. It can still happen that two participants start before they have heard anything from the other one. Since they are listening, they will notice that what they receive/hear is not identical to what they are sending/talking. That means there is a conflict, both are obliged to stop, wait a bit, and then restart. If conflicts repeat, they will wait twice as long next time, etc, until the message gets through.
The effect is a very efficient use of a single high-speed transmission capability. Still, an Ethernet should not be loaded too much. If there is 50% usage, everyone would be delayed about 50% of the time, and actually bit more, due to abandoned and restarted transmissions.
The ethernet protocol is a broadcast protocol, versus a point-to-point protocol, as used by basic telephone sytems. Specifically the Ethernet is a Local Area Network (LAN) using several protocols :
- Carrier Sense Multiple Access (CSMA): the sender first listens to check if it seems ok to put a packet on the line and
- Collision detection (CD): the sender continues to listen, and if the signal (sound) on the line differs from what is being sent, stops and
- Exponential backoff, namely waiting an ever-increasing interval before trying to send again
The technology was developed originally for satellite networks (Aloha net in Hawaii), which transmitted `through the ether', causing long latencies, making it difficult to dynamically schedule allocation of transmission channels for multiple users. (Why was the word ether used for radio transmission?).
Connections to other networks, and eventually to the Internet are acomodated through the Internet protocol (IP). Each Ethernet, or other LAN, will have a router which provides for interconnection of local access (sub) networks (LANs) and larger, Wide Area Networks (WAN). The interfaces must deal with packets that differ according to the specific net requirements in terms of size, speed, and layout. In 22 Nov. 1977 the first IP transmission was demonstrated, among three nets, the mobile net of SRI, operating in a van along US 101, the US satelite network using Aloha mode, and the ARPAnet proper.
VOICE
[rough notes from Goetz Graefe seminar, 11 jan 2008, to be rewritten]
Bob Kahn and Vint Cerf in1974, at Stanford wrote paper.
Asymeyric broadband here.
Dynamic range of Internet is much greater than voice network, which
was predictable by Erlang formulas,
Mobile addressing not supported, needs a layer above, with
authenticity verification.
Other lacunae: broadcast is done by multiple point-to-point connections.
Multihoming with same IP address.
Packeted voice already demonstrated in the SRI van.
Aloha mode first transmitted requirements. Pre-Ethernet.
Reading: CS99 chapter about the Internet.
More (inserted 22 april 2005)
=================================================
Back to What Is The Internet
Back to the CS73N Spring 2008 Schedule
Back to the CS73N Home
