Chapter 3 networking and internetworking

1

3 Networking and Internetworking

As an infrastructure for DS– Distributed computing rely on existing networks: LANs, MANs, WANs

(including internetworks) that use wired and/or wireless technologies

– Hence such characteristics as: performance, reliability, scalability, mobility, and QoS of DS are impacted by the underlying network technology and the OS

Principles of computer networking– Every network has:

An architecture or layers of protocols Packet switching for communication Route selection and data streaming

2

Networking and Internetworking – 3.1 Intro

Comm Subsystems (network technologies rest on):– Transmission media: wires, cables, fiber, wireless (sat, IR, RF, µwave)– Hardware devices: routers, switches, bridges, hubs, repeaters,

network interfaces/card/transceivers– Software components: protocol stacks, comm handlers/drivers, OS

primitives, network-focus APIs

Hosts– The computers and end-devices that use the comm subsystem– Subnet: A single cluster or collection of nodes, which reach each other

on the same physical medium and capable of routing outgoing and incoming messages

– The Internet is a collection of several subnets (or intranets)

3


Networking issues for distributed systems– Initial requirements for DS applications: ftp, rlogin, email, newsgroup– Subsequent generation of DS applics.: on-line shared resources– Current requirements: performance, reliability, scalability, mobility,

security, QoS, multicasting

Performance– Key: time to deliver unit(s) of messages between a pair of

interconnected computers/devices – point-to-point latency (delay) from sending out of outgoing-buffer and receiving into incoming-buffer. Usually due to software overheads, traffic load, and path selection

– Data transfer/bit rate: speed of data transfer between 2 computers (bps). Usually due to physical properties of the medium.

Message trans time = latency + length/bit-rate

4


Bandwidth vs. bit-rate– The total system bandwidth (volume of data sent and received in a

unit time, e.g., per sec.) is a measure of its throughput– Bit rate or transfer rate is restricted to the medium’s ability to

propagate individual bits/signals in a unit time– In most LANs, e.g., Ethernet’s, when full transmission capacity is

devoted to messaging (with little or no latency), then bandwidth and bit-rate are same in measure

– Local memory vs network resources: Applications access to shared resources on same network usually under

msec Applications access to local memory usually under µsec (1000x faster) However, for high speed network web-server, with caches, the access

time is much faster (than local disk access due to hard disk latency)

5


Scalability (Internet and DSs)– Future growth of computing nodes of Internet (hosts, switches) in 109’s (100’s

of 106 hosts alone)– Requires substantial changes to routing and addressing schemes (more later!)– Current traffic (load) on Internet approx. measured by the latencies (see www.

mids.org), which seem to have reduced (with advances in medium and protocol types).

– Future growth and sustainability depend on economies of use, charge rate, locality/placement of shared resource

Reliability– Failures are typically, not due to the physical medium, but at the end-end (at

host levels) software (application-level), therefore, error detection/correction is at the level

– Suggesting that the communication subsystem need not be error-free (made transparent/hidden to user) because reliability is somewhat guaranteed at the send/receiver ends (where errors may be caused by, e.g., buffer overflow, clock drifts causing premature timeouts)

http://www.mids.org/



6


Security– Most intranets are protected from external (Internet-wide) DSs by firewall

– A firewall protects all the resources of an organized from unlawful/malicious access by external users, and control/monitoring of use of resources outside the firewall

– A firewall (bundle of security software and network hardware) runs on a gateway – the entry/exit point of the corporate intranet

– A firewall is usually configured based on corporate security policy, and filters incoming and outgoing messages

– To go beyond firewalls, and grant access to world- or Internet-wide resources, end-to-end authentication, privacy, and security (Standards) are needed to allow DSs to function

– E.g., techniques are Cryptographic and Authentication – usually implemented at a level above the communication subsystem

– Virtual Private Network (VPN) security concept allows intranet-level protection of such features/devices as local routers and secure links to mobile devices

7


Mobility– Need wireless to support portable computers and hand-held devices

– Wireless links are susceptible to, e.g., eavesdropping, distortions in medium, out-of-sight/range transmitters/receivers

– Current addressing and routing schemes are based on ‘wired’ technologies, which have been adapted and, therefore, not perfect and need extensions

QoS (Quality of Service)– Meeting deadlines and user requirements in transmitting/processing streams

of real-time multimedia data

– E.g., QoS requirements: guaranteed bandwidth, timely delivery or bounded latencies, or dynamic readjustments to requirements (more later in Chp 15)

8


Multicasting– Most transmissions are point-to-point, but several involve one-to-many (either

one-to-all – broadcast or selective broadcast – multicast)

– Simply sending the same message from one node to several destinations is inefficient

– Multicasting technique allows single transmission to multiple destination (simultaneously) by using special addressing scheme

9

Networking and Internetworking – 3.2 Type of Networks

3.2 Types of Networks– LANs: (confined to smaller, typically, 2.5km diameter spread)

higher speed, single medium for interconnection (twisted pair, coax, opt), no routing within ‘segments’ – all point-to-point (from hub), inter-segment connections via switches/hubs, low latency, low error rate

E.g., Ethernet, token ring, slotted ring protocols, wired. (1) Ethernet: 1970 with bandwidth of 10Mbps, with extended versions of 100/1000Mbps, lacking latency and bandwidth QoS for DSs: (2) ATM – using frame cells and optical fills the gap but expensive for LAN, newer high-speed Ethernets offer improvement and cost-effective

– MANs: (confined to extended, regional area, typically, up to 50km spread)

Based on high-bandwidth copper and fiber optics for multimedia (audio/video/voice),

E.g., technologies: ATM, high-speed Ethernet (IEEE 802.6 – protocols for MANs), DSL (digital subscriber line) using ATM switches to switch digitized voice over twisted pair @ 0.25-6Mbps within 1.5km, cable modem uses coax @ 1.5Mpbs using analog signaling on TV networks and longer distances than DSL

10


– WANs: (worldwide, lower speeds over sets of varying types of circuits with routers) High latency (due to switching and route searching) between 0.1-0.5s, signaling

speed around 3x105km/s (bounds latency) plus propagation delay (round-trip) of about 0.2s if using satellite/geostationary dishes; generally slower at 10-100kbps or best 1-2Mbps

– Wireless: (connecting portable, wearable devices using access points) Common protocol – IEEE 802.11 (a, b, and now g) (WaveLAN) @ 2-11Mbps (11g’s

bandwidth near 54Mbps) over 150m creating a WLANs, some mobiles connected to fixed devices – printers, servers, palmtops to create a WPANs (wireless personal area networks) using IR links or low-powered Bluetooth radio network tech @ 1-2Mbps over 10m.

Most mobile cell phones use Bluetooth tech. e.g., European GSM standard and US, mostly, analog-based AMP cellular radio network, atop by CDPD – cellular digital packet data communication system, operating over wider areas at lower speed 9.6-19.2kbps.

Tiny screens of mobiles and wearables require a new WAP protocol

– Internetworks Building open, extendible system for DSs, supporting network heterogeneity, multi-

protocol system involving LANs, MANs, WLANs, connected by routers and gateways with layers of software for data and protocol conversions – creating a ‘virtual network’ using underlying physical networks

E.g., the Internet using TCP/IP (over several other physical protocols)

11


–Comparisons– Range of performance characteristics:– Frequency and types of failures, when used for DS applics– Packet delivery/loss, duplicates (masked at TCP level to guarantee some reliability and transparency to DSs; but may use UDP – faster but less reliable and DS applic’s responsibility to guarantee reliability)

12

Networking and Internetworking – 3.3 Network Principles

3.3 Network Principles• Packet Transmission

• Packet transmission superseded telephone/telegraph switched network• Messages are packetized and packets are queued, buffered (in local storage), and transmitted when lines are available using asynchronous transmission protocol

• Data Streaming• Multimedia data can’t be packetized due to unpredicted delays. AV data are streamed at higher frequency and bandwidth at continuous flow rate• Delivery of multimedia data to its destination is time-critical / low latency – requiring end-to-end predefined route• E.g. networks: ATM, IPv6 (next generation – will separate ‘steamed’ IP packets at network layer; and use RSVP (resource reserv. protocol) resource/bandwidth prealloc and RTP play-time/time-reqs (real-time transp protocol) at layers 3 & 1, respectively) to work

13


Switching Schemes – 4 Kinds of switching methods typically used

– Broadcast – no switching logic, all nodes ‘see’ signals on circuits/cells (e.g., Ethernet, wireless networks)

– Circuit Switching – Interconnected segments of circuits via switches/exchange boxes, e.g., POTS (Plain Old Telephone System)

– Packet Switching – Developed as computing tech advanced with processors and storage spaces using store-and-forward algorithms and computers as switches. Packets are not sent instantaneously, routed on different links, reordered, may be lost, high latency (few µsec – msecs). Extension to switch audio/video data brought integration of ‘digitized’ data for computer comm., telephone services, TV, and radio broadcasting, teleconferencing

– Frame Relay – PS (not instantaneous, just an illusion!), but FR, which integrates CS and PS techniques, streams smaller packets (53 byte-cells called frames) as bits at processing nodes. E.g., ATM

14


• Protocols –• Protocols – implemented as pairs of software modules in send/receive nodes,

• Specify the sequence of messages for transmission• Specify the format of the data in the messages

• Protocols Layers – layered architecture, following the OSI suite• packets are communicated as peer-to-peer transmission but effected vertically across layers by encapsulation method over a physical medium

15


Protocols Layers – layered architecture, following the OSI suite

• each protocol type is included in headers to help protocol stack at receiver end to unpack the encapsulated packets

16


Protocols Suites – The 7-layered architecture of the ISO-OSI • Each layer provides service to the layer above it and extends the service provided by the layer below it

• A complete set of protocol layers constitute a suite or stack• Layering simplifies and generalizes the software interface definitions, but costly overhead due to encapsulations and protocol conversions

17


18


19


Protocols

• Packet Assembly:Decomposing messages (packetizing) into packets, transmitting, and reassembling

using sequence #s at delivery-switch to receiving host in the transport layer. Applied to messages that exceed MTU (Max. transfer unit) of the switch. E.g., Ethernet MTU is 1518 bytes and Internet MTU is 8kbyes (min) to 64kbytes (max).

• Ports:Software-defined transmission/delivery points for network-independent transport service on a host computer. Processes are typically attached to ports for pair-wise communication

20


ProtocolsAddressing: Transport layer addressing scheme, composed of network address (of host), I.e., the

IP address, and the port number. The combined address is typically called a socket or transport address of the Transport Layer. Each host may have several port #s for different kinds of protocols (e.g., for HTTP, FTP) or services. Hosts send port numbers to clients to establish, e.g., TCP, connection. Finding port # on server hosts in DS for arbitrary services requires RMI/RPC type of schemes

• Packet Delivery (at network layer): • Datagram – one-at-a-time, hop-by-hop transmission of packets with no storing of

copies at switches, no setup of paths, unreliable and failures are handled by hosts, each packet contains full network address of source-to-destination, e.g., Internet IP datagram in network layer and some wireless networks

• Virtual circuits – set up of end-to-end path/address held in switch tables, no network address in packets except VC #, switching at intermediate nodes, more reliable, latency depends on time to use the links/path segments, unlike POTS voice-links VC links can be shared and used/entered in multiple tables, e.g., ATM[Note: At transport layer, connection-oriented TCP is like virtual circuits, and connection-less UDP is like datagram]

21


–Routing–Routing is necessary in MANs and WANs, rarely in LANs since point-to-point is typically used in LANs. Adaptive/dynamic routing is usually used – adapting to traffic patterns, topological changes, etc. Switching is done by multiple switches/routers in the subnet for host-to-host delivery using available routing algorithm.–Algorithms depends on: 1) Either using VC or datagram - depends on network type, e.g., ATM uses VC connection-oriented and Internet uses datagram connectionless packet-switching; and 2) dynamics of the network – topologically, traffic patterns–Routing decision is made hop-by-hop, with period update and distribution of traffic data, e.g., the distance-vector, dynamic, distributed algorithm

22


The Routing Table – matrix/graph construction, reflecting topology of network

23


The RIP algorithm for dynamic update and distribution of routing table info:

• Prepare RIP packets containing change-info and send to active links and update table if the new cost to a neighboring node is lower/better

24


Congestion Control– Link overload and queue overflows

– Packet dropping – manageable at network layer using retransmission up to a threshold/limit (when throughput starts to decline)

– Congestion control methods arrest overload problem early (at higher nodes – closer to hosts) or buffering of packets for longer times at intermediate nodes, or hosts throttle application programs and/or queue packets in hard-drives –

– Example: In datagram/IP/Internet connectionless networks, where host is responsible for

network problems, choke packets are used to throttle senders In ATM, using connection-oriented protocol, congestion control schemes depend on

the QoS specified in the service

25


Internetworking– Network technologies (or subnets):

LANs: Ethernet, ATM networks using different physical, data link, and network layers

WANs: Internet, using analog and digital POTS switched technologies, satellite links and wide-area ATM networks, and relying on underlying LANs and MANs

– Internetworking: Integrated network of subnets using

• 1) unified internetworking addressing scheme for communication between host and any subnet

• 2) PDU (protocol data unit) format and conversion/handling protocols• 3) standards/protocols and devices/switches for interconnecting and

addressing component subnets and hosts

Network (hardware) components: routers, bridges, hubs, switches Tunneling: Internetworking protocol, e.g., IPv6, for bridging a variety of physical

subnets using ‘packet encapsulation’ techniques. E.g., IPv6 protocol packets encapsulated inside IPv4, IP, ATM PDU’s and transported across a sea of IPv4, IP, ATM networks. Another, e.g., MobileIP transmits IP packets to other mobiles by encapsulating IP packets over other networks, Another, e.g., PPP for transmitting IP packets.

26


27


28

Networking and Internetworking – 3.4 Internet Protocols

29


Internet Protocols– History: 1970’s research results. TCP – Transport control protocol, IP –

Internet protocol

– Forms a single ‘internetworking’ protocol (using IP datagram ‘encapsulation’ methods)

– Many existing application-specific/layer protocols are based on / using TCP/IP i.e., built on top of TCP/IP – (e.g., Web (HTTP), SMTP, POP, FTP, Telnet)

– When TCP is not enough additional higher-level protocol, e.g., SSL (secure socket protocol) for security, can be built atop TCP

– Internet protocols were initially developed for simple ftp and e-mails– Exceptional networks not using TCP/IP – WAP and protocols for multimedia

– Internet protocols usually layered over existing ‘physical’ networks, e.g., over Ethernets and over telephone serial lines via PPP for modem connection

30


– Encapsulation

‘Tags’ in the encapsulation help in determining and conversion (packing / unpacking packets) among protocol types

31


Conceptual (user view) architecture of TCP/IP over transmission networks

32


33


34


35


36


37


38

Networking and Internetworking – 3.5 Network case studies

39


40


41


42


Chapter 3 networking and internetworking

Documents

access time

transfer rate

physical medium

underlying network technology

charge rate

throughputbit rate

local disk access

high speed network webserver