T1: Network Layers: App.,Trans.,Net.,Link,Phys| Networks divided to Layers. Standard Protocols: HTTP, TCP, IP|Protocol: Rules Governing communication between network devices| Devices comm. by req+resp. messages| Protocols specify syntax, semantics, timing of req.+resp.
messages| Syntax: Format of msg| Semantic:Meaning of header field values| Timing: What end sys should do and when Before comm.: TCP Handshake: 1) conn. req. msg.: Establish conn. w server 2) conn. resp. msg.: 3) comm. : Client asks for webpage, server replies with content
of page Layers: Grouping of similar funct. Each layer defines+runs its own prot. Service Model: Layer i gives service layer i+1, layer i gets service from layer i-1| They are modular – easier to impl. change in one layer w/o affecting another
Internet Model: ATNLP| OSI Ref: APSTNLP| Appl (AL): provides interface to user, Present.: Data rep., encryption, compression.| Session: Session setup, teardown, synch. of data exchange (checkpoint + recovery), Trans. (TL): provides AL with reliable service, transports pkts. from src.
to dest. applications, Network (NL): Moves pkt. from src. host to dest. host in diff. networks, Link: deliver PDU from 1 network to other., move pkt. from 1 node to adj. node, Phys.: convert PDU bits into physical signals | Msg. sent between src.+dest. Hosts| Internet msg. broken into
pkts. Internet=Pkt. Switching Network. | A layer in src. host is “peer” to same layer in dest.| Peer Layers: same language, identical packet formats, logically connected. PDU: pkts. aka Protocol Data Units. = Header + Payload + Trailer.| Header+ Trailer = Overhead| P has data, T not
always there| P: Imp. Data to be delivered| H: overhead needed to deliver payload, has src.+dest. addr. | AL PDU=msg., TL PDU=segment, NL PDU=Datagram, LL PDU=Frame, PL PDU = Bits| End Sys. (PCs) =L5 (Application) ,L4 (Transport), Edge Routers=L3 (Network), Switch=L2
(Link), Hub=L1 (Physical) | Protocol Efficiency = Lp/(Lp+Lh), Lp = Payload, Lh = Header| OH = Header+Trailer When OH down, eff. up, when OH up, eff. down| If H close to P, also nor good. | Internet = LAN+WAN| Network Edge: End Systems, set of these form LAN, Network Core:
Routers belonging to ISPs, set of them form WAN | T2: LAN: Set of hosts attached to one router interface. Router Interface is gateway router of LAN End Sys.| Ip add. = NetID + HostID| NetID for LAN, HostID for host in LAN
Hosts of a LAN have same NetID == NetID of Gateway Router.| A LAN has: End Sys., Access Switches (traffic whats in Core Sys.), Edge Router: connect LAN to ISP.| Topologies: Star: Bi: Nl=Nh, Uni: Nl=2Nh| Mesh: Provides direct+dedicated link between any 2 nodes, Bi: Nl=1/2Nh(Nh-
1), Uni: Nl=Nh(Nh-1)| P2P: 2 sys. only, Bus: All sys. attached to one broadcast medium, Ring: All sys attached in seq., Star: All sys attached to 1 central switch, Mesh: All sys are interconnected to each other|
WAN: collection of Internet Core Routers.| WAN Interface: port that faces other routers in public internet| WAN is either point to point OR switched| P2P: consumer grade DSL Link, Switched: enterprise-grade link.| AL=Names identifying end sys., TL=Port No. identify running apps,
NL=IP add. Identify end sys., LL=MAC add. in NIC adapter. Encapsulation from AL msg, TL, NL, LL = [Hl, Hn, Ht, M].| Encap.+Decap happens in intermediate nodes.
Datagram Pkt. Switching: -connectionless: No connection setup before data exch, bandwidth is shared, not dedicated., switches called routers determine best route| Dis. Adv.: random queueing delay, out of order, lost before arrival, bit error after arrival, Protocol Eff. reduced cause
dest. add. carried in pkt. header| Virtual Circuit: Has 2 phases: setup + teardown| Setup: src. + dest. add. of sender+receiver used to make table entries for service| Teardown: src. + dest. inform router to delete correspond. entries| Pkt. will have src. + dest. addr. But also contains
flow label called VC identifier that defines virtual path to follow| Used in connection-oriented WANs, -no queueing delay, no out of order, pkts rarely lost, might have bit error.
T3: Network apps run on end sys.| 2 types of apps: Client and server programs| Client Server: Client: Not always on, always initiates connection, has temp. add.| Server: Permenant add., always on, waits for connection amd accepts request from client.| Peer to Peer Arch.: No reliance
on dedicated servers, high self-scalability, security+performance challenges| AL is run between end sys. at network edge NOT in core| Client AL=consumer devices, Server AL=servers in Datacenters.| Network sockets: Software IF between AL and TL process.| Socket add. = (IP add.
(unique for each host), Process Port no.(Process identified in host)| TL Services to AL: RDT Service=In order, lossless, error free, Security=Confidential+user auth., Qos=guarantees certain max network delay+min successful transmission rate.| Loss-intolerant: Apps that need RDT|
Bandwidth-sensitive: Apps that req. min throughput| Delay-sensitive: Apps that req. small delay| Nodal Delay for 1 pkt.= Tpr+Tq+Ttr+Tprop, Ttr = Lp/Rb, Tprop = d/c, wireless = 3x10^8, wired (optical) = 2x10^8.| Tprop due to hdr. Proc. And error checking = negligible| tq =
Ttr p/(1-p), where p = λ/μ, where λ is pkt. arrival rate, μ is pkt. dep. rate. Delay Jitter: Variation in queueing delays from pkt. to pkt.| Throughput = rate of successful data transmission | TCP vs UDP: UDP: connectionless, data sent immed. From C to S, Not RDT so no guarantee for
successful transfer, fine for apps that tolerate data loss| TCP: connection oriented, is RDT so in order, lossless, errorfree.| TCP with security: SSL| Both: NOT secure-No encryption, run on IP protocol that’s best-effort, cant provide throughput/delay guarantees
T4: Email comp.: User Agents: Software/Browser, Mail Server: Sending, receiving, storing, Protocols: SMTP: Push prot., POP/IMAP: Pull prot.| Emails require RDT from TL., uses TCP.| SMTP: uses client server, receives on port 25, C and S perform AL level handshake, persistent
connection: If several messages to 1 server, can send all in one connection Mail server could be SMTP client OR server. Email msg. is in 7 bit ASCII Format, if is non ASCII, must be encoded to 7-bit ASCII| POP vs IMAP: POP: msg. storage limited by PC capacity, if msg. on PC it could be
lost if PC fails, If read email from diff devices, emails in diff places.| IMAP: Msg. in server, can use multiple PC, storage limited by PC Capacity| MIME: mail extension for non-ASCII multimedia| In content header, content type + encoding specified IANA resp. for all MIME types| base64
encodes binary data like JPEG to ASCII| NetID tells us which network a host belongs to.| T DNS translates names to IP add.(is 4 bytes), AL protocol: uses C S, runs on UDP/TCP port 53, database on server| DNS client: sends query msg., DNS server: sends reply msg.| msg.: 12 bytes,
header has 16 bit ID and Flags.| Server hierarchy: Root->TLD->Auth.->Local| Records are distributed cause single DNS server is single point of failure and cant be scaled to billions of reqs.| Root: Gives IP of TLD server, many physical machines: reliable+scalable| TLD+Auth.: TLD:
provides IP for Auth. Like .com (Verisign), .edu (Educause)| Auth: maintains public DNS record of Web+Mail Server. (All aus.edu)| Local DNS = Default Name Server, not part of hierarchy, fast reply, queries first sent to local DNS and forwarded to hierarchy, can also cache IP
(discarded after 2 days) of TLD (bypassing root server), Used by ISP, firms, unis|
T5: HTTP: Browser sends request, web server sends response. Uses TCP, requires RDT from TL| Client=Browser, Server=Apache Server. (TCP Port 80)| Web Page Comp.: Webpage: base HTML + referenced objects with URLs. | URL (Uniform Resource Locators) has Hostname +
Object Path| Web Cookies: Initially HTTP was stateless – focused on performance in mind| “Stateful” HTTP = HTTP + Cookies as can identify returning users.| 4 comps.: Has cookie header line in Req., same in Resp. msg., cookie file in user end sys and backend database at web server
location.| HTTP Delay: Delay of 1 Obj using RTT: T[obj] = 2RTT + Ttr[obj]| Persistent: multiple objects can be sent in single TCP connection| Pipelined: Client sends request as soon as it encounters referenced obj.| Store and Forward: Node must receive the entire pkt. before
transmitting it to the next node. HTTP 1.0: non persistent + non pipelined, T[N obj] = N(2RTT+Ttr)| HTTP 1.1: persistent and can be pipelined, w/o pipeline: T[N obj] = RTT + N(RTT+Ttr), with pipeline: T[N obj]=2RTT + N Ttr| HTTP 2: Default for most, faster+more secure than 1,
server can “push” content, no need for requesting, encrypted with HTTPS as default.| HTTP3: Fixes HOL(head of line) Blocking Problem of HTTP 2, HOL: 1 lost TCP pkt. causes delays for all other active conn.
T6: E2E delay, with N links, N-1 Routers, 1 pkt.: Te2e[1 pkt.]=N(Ttr + Tprop) + (N-1)(Tq+Tpc)| E2E for M pkts, N links, N-1 Routers: (M)Ttr + (N-1)(Tpc+Tq+Ttr)+(N)Tprop OR [M+N-1]Ttr + (N-1)(Tpc+Tq)+Tprop|Pkt. loss reasons: Noisy links cause bit err./pkt.
err., Routers have finite buffer capacity, if full buffer, pkt lost,or AQM to prevent cong.| Throughput affected by: network congest., Bit error rate + pkt. loss/error, pkt dropping by routers.| Noise increases prob. Of bit error (BER) Throughput to bit errors = Rb, Ps = Rb(1-Pb)^L, L = bits
in pkt., Pb = bit error rate, Ps is prob. that pkt. is error free| Caching stored in local storage and/or proxy server called institutional cache server.| Cache servers allow HTTP request w/o involving origin servers.| Faster resp. time, Reduces traffic intensity, could be out dated = solution
is timestamp + conditional GET method. Cache server stores “last modified date” and GET has field “if modified since” If server has newer version of object, then GET object, else don’t. RTT w/o web cache: Ttot = Tlan+Tacc+Tint (each has req and resp.)| w web cache Tavg for all |
req. = ⴄTlan+(1- ⴄ)(T’acc+Tlan+Tint), p’=(1- ⴄ )p -> [access link new traffic intensity] | req./resp. rate inside LAN = λ, outside LAN = (1- ⴄ)λ| TLAN = TreqLAN + TrespLAN = [treqtr+treqq+trsptr+trspq]LAN| TLAN = TreqACC + TrespACC = [treqtr+treqq+trsptr+trspq]ACC| T10: NL support for MM
apps. is difficult: stringent QoS requirements, network congestion. Proper Network Dimensioning: Provide enough capacity everywhere in best-effort network = prevents congestion| Bandwidth provisioning (how much link capacity needed for network topology) + network
dimensioning (how to design network topology to achieve a e2e performance) needed| CDN: Overlaying infrastructure over best-effort networks| Determine which server is best when a client req. content, performs content processing for decent quality, uses AL techniques like
adaptive playback and buffering aka DASH to mitigate issues| Multiple service classes: Used by IETF Diffserv favors some classes over others but can’t guarantee Qos, uses mult. Classes of service: P1: Pkt. classification + marking, P2: Scheduling, P3: Metering + Policing| P1: Mark
H1-H3 VoIP pkts. as VIP with black mark. Mark H2-H4 pkts. as reg., non-VIP pkts. + grey mark| P2: 2 Scheduling: FIFO: useful if have pkt. discarding policy, Priority: Schedule from highest priority as long as queue not empty, good for VIP pkts. but harsh on normal pkts. Weighted Fair
queueing is good compromise: schedules pkts. from all traffic, long trans. times are higher priority| P3: Metering measures what sources are actually transmitting into link: compares incoming pkt. flow with agreed traffic profile| Policing: forces sources to adhere with some
parameters, alg. like leaky bucket used to shape traffic. With both WFQ and leaky bucket, upper bound of delay achieved. Avg. rate: limits pkt. rate that can be sent in LR, Peak Rate: limits max pkt. rate that can be sent in short periods, Burst Size: limits max no. of pkts. sent
consecutively w/o idle time. T9: MM = digital text, audio, video, images.| Text: chars encoded with ASCII or UTF (symbols), Image: Matrix of pixels, Video: sequence of frame, each frame = 1 image, needs high Ttr and compression, Audio: voice+music is digitized analog signals,
humans 20Hz – 20khz| MM application types: Streaming live audio/video: No interaction, cannot prefetch, limits effectiveness of buffering, (Least strict) Streaming stored audio/video: start playout while file downloading, requires buffering to absorb delay jitter| (Most strict) Interactive
audio/video: limits delay tolerance, most demanding/ stringent., real time human comm.| Text: ASCII encoding for alphanum, special char and ctrl char. Unicode for other lang. lossless text compression: Lempel-Ziv-Welch(LZW) 5:1 compression ratio, does NOT tolerate lossy comp.|
Audio: analog converted to digital using PCM, sampled at constant rate sf(sampling freq.) higher freq. of sample = more sampling must happen, telephony = 8000 samp./s., CD = 44100 samp./s., 8 bits = 2^8 = 256 quantized values, telephony+PCM+w/o comp=64 kbps,
CD+PCM+w/o comp.=1411 kbps| lossy and lossless compression, cell phones (8,10,12 kbps), internet call (5) and mp3 (96,128,160) uses lossy| Image: B&W depth 1 bit, Gray image depth 8 bits, RGB depth 8x3=24 bits.| JPEG uses DCT compression (saves time+bandwidth). High
freq = sudden change in pixel value. Comp. removes high freq. Strong lossy introduces artifacts| Video: compression is Spatial and Temporal Redundancy and MPEG. Spatial: comp. in each image (JPEG img. Comp.), Temporal: comp. between frames (redundant frames removed),
MPEG uses both prev. redundancies and treats frames as GOP (group of pict.) with I frames (JPEG img. indep. From other frames.) and P and B frames(smaller in size, depends on I frame)| Transmission Time: | Video tr rate: pixel h x pixel w x fps x bpc| VoIP Requirements: Low e2e
delay + low pkt loss| delay<150 ms: good, 150<delay<400 is noticeable, impairs interactivity, delay>400: bad/unusable| Loss < 10% can be concealed using FEC, loss > 20% unusable (no solution)| Pkt loss mitigation: Pkt interleaving + pkt loss concealment| Delay Jitter mitigation:
Delayed playback or adaptive playback. Network Delay from graph: between two vertical lines, playout schedule: the blue dot minus the first pkt received vertical line.
T12: Hosts have unique interface IP addresses, host identifier is its IP add., process identifier is its port no., Socket add. identifies process on any host| TL segment header = 16 bits| Port add. are from 0-65,535, well known 0-1023 (for standard server processes), Registered 1024-
49,151 (reg. by IANA to prevent duplication), Dynamic >49,152 (assigned by OS to client processes)| HTTP(80), HTTPS(443), SMPT(25), POP3(110), DNS(53) by IANA for server processes, Dynamic ports reserved for client processes | UDP Error detection: Checksum, UDP doesn’t
recover error, discards seg. With error, don’t know if out of order, don’t know if missing segment| TCP sets up conn. with src. and dest. Processes, Tears down after data transfer, in order delivery even when pkts. delivered are out of order, detects data loss + retransmits lost pkts, If
only pkt. error, T RX uses checksum to detect error, tells TX if good(send ACK)/bad(send NAK) pkt., tells TX what to retransmit using using pkt. seq. no.| If error and loss, use CS+ACK+seq. no. and timer+timeout (to handle loss)| ARQ: Automatic Repeat Request used by TCP. ARQ
Types: non pipelined S&W and pipelined sliding window (SR and GBN)| S&W: TX sends 1 pkt at a time. TX and RX window size = 1, pkt. seq. alternates between 0 and 1, ACK alternates between 1 and 0, ACK Value is the next expected seq. by RX. 1 Timer used for the pkt sent, if TX
does not receive ACK + timer expires, TX retransmits the pkt| Wrong with SW, Link Utilization FOR 1 PKT. = Ttr / (Ttr + RTT) = 1/(1+ap), ap = (RTT/Lp) * Rb = RTT/Ttr, ap is no. of pkts. transmitter in 1 RTT| SW has poor utilization: no pipelining| GBN and SR (both are sliding window)
are pipeline so good link utilization| Link utilization of Sliding Window ARQ Upl = NTtr/(RTT+Ttr) When expected ACK arrives, TX window slides, when expected pkts arrive RX window slides| pkt. sequence is m bit field in p kt. header, total possible seq. = 2^m| GBN: TX window size N
<= 2^(m)-1, Rx window N = 1. TX has 1 timer for oldest outstanding unACKed pkt, when timer expires, TX resends all outstanding pkts. GBN uses cumulative ACK, depending how many ACKs received, it will slide that way, only sends ACK for error-free pkt. and with expected seq.
only. RX handles out of order pkts. by discarding them, doesn’t buffer. | SR: TX window N <= 2^(m-1), RX window = TX window. Every unAcked pkt. has a timer in TX, TX sends only pkt. that’s timer expired. SR does NOT use Cumulative ACK, SR may buffer out of sequence pkts.
until contiguous set formed, if UDP checksum not calc., then field is all 0s | T13: Why UDP?: UDP is best effort, unguaranteed service, Connectionless – No handshaking so less delay, segments handled indep., simple, lightweight small header. No congest. ctrl.: can go as fast as it
wants| DNS is example of AL that uses UDP| UDP segment: Header = 8 bytes Order: src. no., dest. no., length, checksum, msg. Max UDP segment size = 2^16 bytes, max UDP payload size = 2^16 – 8 bytes| Sender checksum, add all and comp. to find, Receiver checksum,
complement final sum, if C= 0, msg. accepted, else not accepted. CS calc. over entire UDP segment and few fields in IP header. T14: TCP: Point 2 point, bi directional, pipelined. TCP Flow control: Sender process at AL is producer (pushes msg to TL). Sender TL is Consumer (consumes
the msg.) and Producer (encaps. msg. to pkts. and push to TL). Receiver TL is also Consumer (receives pkts.) and Producer (decaps. msg and delivers to AL). 2 BUFFERS, one at sending TL and one at receiving TL| Buffers ctrl. flow. If buffer full/close to OF: slow down/stop sending
msgs. If buffer empty/close to starvation: speed up/ can send pkts. again| Flow ctrl. found in 2 places: 1) between sender TCP and sender AL 2) between sender TCP and receiver TCP| Echange Data Process: TX TCP direct data to TX Buffer. Then append header to data = TCP
Segment. Segment send to TX NL. Segment encap. In IP datagram. RX IP gets decap. By RX NL, sent to RX TCP. RX TCP gets segment payload and places in TCP buffer| TCP max segment size (MSS) limited by LL maximum payload called MTU| IEEE 802.3 MTU = 1500 bytes, IEEE
802.11 MTU = 2304 bytes, IEEE 802.5 MTU = 4464 bytes| MSS = MTU – (Htl – Hnl)| standard header size for TCP or IP is 20 bytes| TCP segment structure: src. no., dest. no, seq. no., ack no., HLen, Flags, Rwnd, Checksum, Data Ptr., Options, Data| Hlen is count for no of 4 bit words in
header. Standard is 20 bytes, max = 60 bytes| With ISN = X, MSS = Y, seq no.: X, X+Y, X+2Y…| First byte in Payload is the seg. seq. no| ACK No. = Next seq. no. = current seq No + no. of data (bytes)
T TCP connection: 3 way handshake: Leg 1: SYN=1, ACK=0, Leg 2: SYN=1, ACK=1, Leg 3: SYN=0, ACK=1| TCP teardown: 4 way, teardown can start with C or S. If connection closed in 1 direction, it can still continue other way. C sends FIN = 1 to S, S sends ACK to confirm close, S
continues to send until all send, then sends FIN = 1 back to C, C sends ACK to confirm, timed wait, then closes, wait state is 30s,, 1 min or 2 min| T Cumulative ACK when one ACK is used for that seq. no. and the ones before it, for which the ACK was lost| TCP loss events: Timeout or
Three Dup. ACKs| For TO, when ACK is lost + timer done, then seq resent and ACK received| TDA: when a seq. no is lost, then ACK also lost(say ACK 100) so for next 3 seq. no, the ACK will be 100, so TDA fast retransmit send the pkt. with ACK 100 again| If TO < RTT: Premature TO, If
TO > RTT, Slow reaction to lost segments. TO[𝑘]= RTTa [𝑘] + 4Δa [𝑘] ,Timeout= RTT + safety margin| Let 𝛼 = 0.125, 𝛽 = 0.25 | RTTa[𝑘] = (1−𝛼)RTTa[k-1] + 𝛼RTTi[k]| Δa[𝑘] = (1−𝛽) ⋅Δa [𝑘−1] + 𝛽⋅Δ𝑖[𝑘]| Δi[k] = | RTTa[k] - RTTi[k] | Δa[𝑘] : average deviation | Δ𝑖 : instantaneous
deviation| average values of RTT and a used because instantaneous values have large variances. TCP ARQ Characteristics: pipelined, cumulative, timers. T If no congestion ctrl. then per conn. throughput will collapse, causing: data loss, retransmission, collapse of all connections| 2
approaches for congestion ctrl.: Classic appr. (uses sender perception) And network assisted appr. (network gives explicit feedback to TCP senders) aka ATM approach| Classic appr.: self clocking: TCP sender limits rates (either up or down) depending on perceived network cong.,
done by incr./decr. Cong. Window (Cwnd)| TCP perceives congestion when loss event occurs: either TO: severe, nothing delivered or TDA: Less severe, something delivered| Tahoe: Every RTT a segment is transmitted| Phases: Slow Start (SS): Cwnd is 1 MSS, doubles until initial
threshold, then enters Cong. Avoidance (CA): Cwnd + 1 every time| If TDA or TO occurs, Cwnd = 1, then SS starts and New threshold = ½ Cwnd[old], where Cwnd[old] is Cwnd before fall to 1| Reno: If TO: falls to 1, then SS, If TDA, enters Fast Recovery, Cwnd = ½ Cwnd[old] + 3
MSS, no SS, direct CA| Cwnd decrease is multiplicative| For Reno: Ravg = 1.5 (Wmin / RTT) | New approach - ECN: 2 bits in IP header signal congestion, 11 signals cong. Sent to Host B, then Host B echoes cong. Back to A by ECN Echo Flag ECE = 1, then Host A halves Cwnd and tells
Host B by TCP Header CWR = 1 | Segment Size = Min(Flow Window ctrl., cong. Ctrl. window) = min(Rwnd, Cwnd)| Fairness: UDP is unfair as no congestion ctrl| TCP is fair if assume that TCP connections all have same RTT. TCP Cong. Ctrl. is called additive increase and multiplicative
decrease or AIMD. Reno used AIMD HW3: is a technique that a client-side application can use to mitigate network delay jitter. buffering. Packet interleaving is a common AL technique used to mitigate packet loss in multimedia applications.
The main reason for streaming video files from online servers instead of downloading them is their large sizes. Interactive streamingaudio/video describes the type of multimedia applications such as Skype.
Packet Loss is one of the imporant issues that multimedia applications must address in today's Internet.
To support streaming multimedia content over the Internet, service providers often use CDN overlay architecture.
Delayed Playback is a technique that a multimedia application can use to mitigate network delay.
Header = 12 bytes
UDP Id: 16-bit no. to identify query, copied to reply message
of query.
messages| Syntax: Format of msg| Semantic:Meaning of header field values| Timing: What end sys should do and when Before comm.: TCP Handshake: 1) conn. req. msg.: Establish conn. w server 2) conn. resp. msg.: 3) comm. : Client asks for webpage, server replies with content
of page Layers: Grouping of similar funct. Each layer defines+runs its own prot. Service Model: Layer i gives service layer i+1, layer i gets service from layer i-1| They are modular – easier to impl. change in one layer w/o affecting another
Internet Model: ATNLP| OSI Ref: APSTNLP| Appl (AL): provides interface to user, Present.: Data rep., encryption, compression.| Session: Session setup, teardown, synch. of data exchange (checkpoint + recovery), Trans. (TL): provides AL with reliable service, transports pkts. from src.
to dest. applications, Network (NL): Moves pkt. from src. host to dest. host in diff. networks, Link: deliver PDU from 1 network to other., move pkt. from 1 node to adj. node, Phys.: convert PDU bits into physical signals | Msg. sent between src.+dest. Hosts| Internet msg. broken into
pkts. Internet=Pkt. Switching Network. | A layer in src. host is “peer” to same layer in dest.| Peer Layers: same language, identical packet formats, logically connected. PDU: pkts. aka Protocol Data Units. = Header + Payload + Trailer.| Header+ Trailer = Overhead| P has data, T not
always there| P: Imp. Data to be delivered| H: overhead needed to deliver payload, has src.+dest. addr. | AL PDU=msg., TL PDU=segment, NL PDU=Datagram, LL PDU=Frame, PL PDU = Bits| End Sys. (PCs) =L5 (Application) ,L4 (Transport), Edge Routers=L3 (Network), Switch=L2
(Link), Hub=L1 (Physical) | Protocol Efficiency = Lp/(Lp+Lh), Lp = Payload, Lh = Header| OH = Header+Trailer When OH down, eff. up, when OH up, eff. down| If H close to P, also nor good. | Internet = LAN+WAN| Network Edge: End Systems, set of these form LAN, Network Core:
Routers belonging to ISPs, set of them form WAN | T2: LAN: Set of hosts attached to one router interface. Router Interface is gateway router of LAN End Sys.| Ip add. = NetID + HostID| NetID for LAN, HostID for host in LAN
Hosts of a LAN have same NetID == NetID of Gateway Router.| A LAN has: End Sys., Access Switches (traffic whats in Core Sys.), Edge Router: connect LAN to ISP.| Topologies: Star: Bi: Nl=Nh, Uni: Nl=2Nh| Mesh: Provides direct+dedicated link between any 2 nodes, Bi: Nl=1/2Nh(Nh-
1), Uni: Nl=Nh(Nh-1)| P2P: 2 sys. only, Bus: All sys. attached to one broadcast medium, Ring: All sys attached in seq., Star: All sys attached to 1 central switch, Mesh: All sys are interconnected to each other|
WAN: collection of Internet Core Routers.| WAN Interface: port that faces other routers in public internet| WAN is either point to point OR switched| P2P: consumer grade DSL Link, Switched: enterprise-grade link.| AL=Names identifying end sys., TL=Port No. identify running apps,
NL=IP add. Identify end sys., LL=MAC add. in NIC adapter. Encapsulation from AL msg, TL, NL, LL = [Hl, Hn, Ht, M].| Encap.+Decap happens in intermediate nodes.
Datagram Pkt. Switching: -connectionless: No connection setup before data exch, bandwidth is shared, not dedicated., switches called routers determine best route| Dis. Adv.: random queueing delay, out of order, lost before arrival, bit error after arrival, Protocol Eff. reduced cause
dest. add. carried in pkt. header| Virtual Circuit: Has 2 phases: setup + teardown| Setup: src. + dest. add. of sender+receiver used to make table entries for service| Teardown: src. + dest. inform router to delete correspond. entries| Pkt. will have src. + dest. addr. But also contains
flow label called VC identifier that defines virtual path to follow| Used in connection-oriented WANs, -no queueing delay, no out of order, pkts rarely lost, might have bit error.
T3: Network apps run on end sys.| 2 types of apps: Client and server programs| Client Server: Client: Not always on, always initiates connection, has temp. add.| Server: Permenant add., always on, waits for connection amd accepts request from client.| Peer to Peer Arch.: No reliance
on dedicated servers, high self-scalability, security+performance challenges| AL is run between end sys. at network edge NOT in core| Client AL=consumer devices, Server AL=servers in Datacenters.| Network sockets: Software IF between AL and TL process.| Socket add. = (IP add.
(unique for each host), Process Port no.(Process identified in host)| TL Services to AL: RDT Service=In order, lossless, error free, Security=Confidential+user auth., Qos=guarantees certain max network delay+min successful transmission rate.| Loss-intolerant: Apps that need RDT|
Bandwidth-sensitive: Apps that req. min throughput| Delay-sensitive: Apps that req. small delay| Nodal Delay for 1 pkt.= Tpr+Tq+Ttr+Tprop, Ttr = Lp/Rb, Tprop = d/c, wireless = 3x10^8, wired (optical) = 2x10^8.| Tprop due to hdr. Proc. And error checking = negligible| tq =
Ttr p/(1-p), where p = λ/μ, where λ is pkt. arrival rate, μ is pkt. dep. rate. Delay Jitter: Variation in queueing delays from pkt. to pkt.| Throughput = rate of successful data transmission | TCP vs UDP: UDP: connectionless, data sent immed. From C to S, Not RDT so no guarantee for
successful transfer, fine for apps that tolerate data loss| TCP: connection oriented, is RDT so in order, lossless, errorfree.| TCP with security: SSL| Both: NOT secure-No encryption, run on IP protocol that’s best-effort, cant provide throughput/delay guarantees
T4: Email comp.: User Agents: Software/Browser, Mail Server: Sending, receiving, storing, Protocols: SMTP: Push prot., POP/IMAP: Pull prot.| Emails require RDT from TL., uses TCP.| SMTP: uses client server, receives on port 25, C and S perform AL level handshake, persistent
connection: If several messages to 1 server, can send all in one connection Mail server could be SMTP client OR server. Email msg. is in 7 bit ASCII Format, if is non ASCII, must be encoded to 7-bit ASCII| POP vs IMAP: POP: msg. storage limited by PC capacity, if msg. on PC it could be
lost if PC fails, If read email from diff devices, emails in diff places.| IMAP: Msg. in server, can use multiple PC, storage limited by PC Capacity| MIME: mail extension for non-ASCII multimedia| In content header, content type + encoding specified IANA resp. for all MIME types| base64
encodes binary data like JPEG to ASCII| NetID tells us which network a host belongs to.| T DNS translates names to IP add.(is 4 bytes), AL protocol: uses C S, runs on UDP/TCP port 53, database on server| DNS client: sends query msg., DNS server: sends reply msg.| msg.: 12 bytes,
header has 16 bit ID and Flags.| Server hierarchy: Root->TLD->Auth.->Local| Records are distributed cause single DNS server is single point of failure and cant be scaled to billions of reqs.| Root: Gives IP of TLD server, many physical machines: reliable+scalable| TLD+Auth.: TLD:
provides IP for Auth. Like .com (Verisign), .edu (Educause)| Auth: maintains public DNS record of Web+Mail Server. (All aus.edu)| Local DNS = Default Name Server, not part of hierarchy, fast reply, queries first sent to local DNS and forwarded to hierarchy, can also cache IP
(discarded after 2 days) of TLD (bypassing root server), Used by ISP, firms, unis|
T5: HTTP: Browser sends request, web server sends response. Uses TCP, requires RDT from TL| Client=Browser, Server=Apache Server. (TCP Port 80)| Web Page Comp.: Webpage: base HTML + referenced objects with URLs. | URL (Uniform Resource Locators) has Hostname +
Object Path| Web Cookies: Initially HTTP was stateless – focused on performance in mind| “Stateful” HTTP = HTTP + Cookies as can identify returning users.| 4 comps.: Has cookie header line in Req., same in Resp. msg., cookie file in user end sys and backend database at web server
location.| HTTP Delay: Delay of 1 Obj using RTT: T[obj] = 2RTT + Ttr[obj]| Persistent: multiple objects can be sent in single TCP connection| Pipelined: Client sends request as soon as it encounters referenced obj.| Store and Forward: Node must receive the entire pkt. before
transmitting it to the next node. HTTP 1.0: non persistent + non pipelined, T[N obj] = N(2RTT+Ttr)| HTTP 1.1: persistent and can be pipelined, w/o pipeline: T[N obj] = RTT + N(RTT+Ttr), with pipeline: T[N obj]=2RTT + N Ttr| HTTP 2: Default for most, faster+more secure than 1,
server can “push” content, no need for requesting, encrypted with HTTPS as default.| HTTP3: Fixes HOL(head of line) Blocking Problem of HTTP 2, HOL: 1 lost TCP pkt. causes delays for all other active conn.
T6: E2E delay, with N links, N-1 Routers, 1 pkt.: Te2e[1 pkt.]=N(Ttr + Tprop) + (N-1)(Tq+Tpc)| E2E for M pkts, N links, N-1 Routers: (M)Ttr + (N-1)(Tpc+Tq+Ttr)+(N)Tprop OR [M+N-1]Ttr + (N-1)(Tpc+Tq)+Tprop|Pkt. loss reasons: Noisy links cause bit err./pkt.
err., Routers have finite buffer capacity, if full buffer, pkt lost,or AQM to prevent cong.| Throughput affected by: network congest., Bit error rate + pkt. loss/error, pkt dropping by routers.| Noise increases prob. Of bit error (BER) Throughput to bit errors = Rb, Ps = Rb(1-Pb)^L, L = bits
in pkt., Pb = bit error rate, Ps is prob. that pkt. is error free| Caching stored in local storage and/or proxy server called institutional cache server.| Cache servers allow HTTP request w/o involving origin servers.| Faster resp. time, Reduces traffic intensity, could be out dated = solution
is timestamp + conditional GET method. Cache server stores “last modified date” and GET has field “if modified since” If server has newer version of object, then GET object, else don’t. RTT w/o web cache: Ttot = Tlan+Tacc+Tint (each has req and resp.)| w web cache Tavg for all |
req. = ⴄTlan+(1- ⴄ)(T’acc+Tlan+Tint), p’=(1- ⴄ )p -> [access link new traffic intensity] | req./resp. rate inside LAN = λ, outside LAN = (1- ⴄ)λ| TLAN = TreqLAN + TrespLAN = [treqtr+treqq+trsptr+trspq]LAN| TLAN = TreqACC + TrespACC = [treqtr+treqq+trsptr+trspq]ACC| T10: NL support for MM
apps. is difficult: stringent QoS requirements, network congestion. Proper Network Dimensioning: Provide enough capacity everywhere in best-effort network = prevents congestion| Bandwidth provisioning (how much link capacity needed for network topology) + network
dimensioning (how to design network topology to achieve a e2e performance) needed| CDN: Overlaying infrastructure over best-effort networks| Determine which server is best when a client req. content, performs content processing for decent quality, uses AL techniques like
adaptive playback and buffering aka DASH to mitigate issues| Multiple service classes: Used by IETF Diffserv favors some classes over others but can’t guarantee Qos, uses mult. Classes of service: P1: Pkt. classification + marking, P2: Scheduling, P3: Metering + Policing| P1: Mark
H1-H3 VoIP pkts. as VIP with black mark. Mark H2-H4 pkts. as reg., non-VIP pkts. + grey mark| P2: 2 Scheduling: FIFO: useful if have pkt. discarding policy, Priority: Schedule from highest priority as long as queue not empty, good for VIP pkts. but harsh on normal pkts. Weighted Fair
queueing is good compromise: schedules pkts. from all traffic, long trans. times are higher priority| P3: Metering measures what sources are actually transmitting into link: compares incoming pkt. flow with agreed traffic profile| Policing: forces sources to adhere with some
parameters, alg. like leaky bucket used to shape traffic. With both WFQ and leaky bucket, upper bound of delay achieved. Avg. rate: limits pkt. rate that can be sent in LR, Peak Rate: limits max pkt. rate that can be sent in short periods, Burst Size: limits max no. of pkts. sent
consecutively w/o idle time. T9: MM = digital text, audio, video, images.| Text: chars encoded with ASCII or UTF (symbols), Image: Matrix of pixels, Video: sequence of frame, each frame = 1 image, needs high Ttr and compression, Audio: voice+music is digitized analog signals,
humans 20Hz – 20khz| MM application types: Streaming live audio/video: No interaction, cannot prefetch, limits effectiveness of buffering, (Least strict) Streaming stored audio/video: start playout while file downloading, requires buffering to absorb delay jitter| (Most strict) Interactive
audio/video: limits delay tolerance, most demanding/ stringent., real time human comm.| Text: ASCII encoding for alphanum, special char and ctrl char. Unicode for other lang. lossless text compression: Lempel-Ziv-Welch(LZW) 5:1 compression ratio, does NOT tolerate lossy comp.|
Audio: analog converted to digital using PCM, sampled at constant rate sf(sampling freq.) higher freq. of sample = more sampling must happen, telephony = 8000 samp./s., CD = 44100 samp./s., 8 bits = 2^8 = 256 quantized values, telephony+PCM+w/o comp=64 kbps,
CD+PCM+w/o comp.=1411 kbps| lossy and lossless compression, cell phones (8,10,12 kbps), internet call (5) and mp3 (96,128,160) uses lossy| Image: B&W depth 1 bit, Gray image depth 8 bits, RGB depth 8x3=24 bits.| JPEG uses DCT compression (saves time+bandwidth). High
freq = sudden change in pixel value. Comp. removes high freq. Strong lossy introduces artifacts| Video: compression is Spatial and Temporal Redundancy and MPEG. Spatial: comp. in each image (JPEG img. Comp.), Temporal: comp. between frames (redundant frames removed),
MPEG uses both prev. redundancies and treats frames as GOP (group of pict.) with I frames (JPEG img. indep. From other frames.) and P and B frames(smaller in size, depends on I frame)| Transmission Time: | Video tr rate: pixel h x pixel w x fps x bpc| VoIP Requirements: Low e2e
delay + low pkt loss| delay<150 ms: good, 150<delay<400 is noticeable, impairs interactivity, delay>400: bad/unusable| Loss < 10% can be concealed using FEC, loss > 20% unusable (no solution)| Pkt loss mitigation: Pkt interleaving + pkt loss concealment| Delay Jitter mitigation:
Delayed playback or adaptive playback. Network Delay from graph: between two vertical lines, playout schedule: the blue dot minus the first pkt received vertical line.
T12: Hosts have unique interface IP addresses, host identifier is its IP add., process identifier is its port no., Socket add. identifies process on any host| TL segment header = 16 bits| Port add. are from 0-65,535, well known 0-1023 (for standard server processes), Registered 1024-
49,151 (reg. by IANA to prevent duplication), Dynamic >49,152 (assigned by OS to client processes)| HTTP(80), HTTPS(443), SMPT(25), POP3(110), DNS(53) by IANA for server processes, Dynamic ports reserved for client processes | UDP Error detection: Checksum, UDP doesn’t
recover error, discards seg. With error, don’t know if out of order, don’t know if missing segment| TCP sets up conn. with src. and dest. Processes, Tears down after data transfer, in order delivery even when pkts. delivered are out of order, detects data loss + retransmits lost pkts, If
only pkt. error, T RX uses checksum to detect error, tells TX if good(send ACK)/bad(send NAK) pkt., tells TX what to retransmit using using pkt. seq. no.| If error and loss, use CS+ACK+seq. no. and timer+timeout (to handle loss)| ARQ: Automatic Repeat Request used by TCP. ARQ
Types: non pipelined S&W and pipelined sliding window (SR and GBN)| S&W: TX sends 1 pkt at a time. TX and RX window size = 1, pkt. seq. alternates between 0 and 1, ACK alternates between 1 and 0, ACK Value is the next expected seq. by RX. 1 Timer used for the pkt sent, if TX
does not receive ACK + timer expires, TX retransmits the pkt| Wrong with SW, Link Utilization FOR 1 PKT. = Ttr / (Ttr + RTT) = 1/(1+ap), ap = (RTT/Lp) * Rb = RTT/Ttr, ap is no. of pkts. transmitter in 1 RTT| SW has poor utilization: no pipelining| GBN and SR (both are sliding window)
are pipeline so good link utilization| Link utilization of Sliding Window ARQ Upl = NTtr/(RTT+Ttr) When expected ACK arrives, TX window slides, when expected pkts arrive RX window slides| pkt. sequence is m bit field in p kt. header, total possible seq. = 2^m| GBN: TX window size N
<= 2^(m)-1, Rx window N = 1. TX has 1 timer for oldest outstanding unACKed pkt, when timer expires, TX resends all outstanding pkts. GBN uses cumulative ACK, depending how many ACKs received, it will slide that way, only sends ACK for error-free pkt. and with expected seq.
only. RX handles out of order pkts. by discarding them, doesn’t buffer. | SR: TX window N <= 2^(m-1), RX window = TX window. Every unAcked pkt. has a timer in TX, TX sends only pkt. that’s timer expired. SR does NOT use Cumulative ACK, SR may buffer out of sequence pkts.
until contiguous set formed, if UDP checksum not calc., then field is all 0s | T13: Why UDP?: UDP is best effort, unguaranteed service, Connectionless – No handshaking so less delay, segments handled indep., simple, lightweight small header. No congest. ctrl.: can go as fast as it
wants| DNS is example of AL that uses UDP| UDP segment: Header = 8 bytes Order: src. no., dest. no., length, checksum, msg. Max UDP segment size = 2^16 bytes, max UDP payload size = 2^16 – 8 bytes| Sender checksum, add all and comp. to find, Receiver checksum,
complement final sum, if C= 0, msg. accepted, else not accepted. CS calc. over entire UDP segment and few fields in IP header. T14: TCP: Point 2 point, bi directional, pipelined. TCP Flow control: Sender process at AL is producer (pushes msg to TL). Sender TL is Consumer (consumes
the msg.) and Producer (encaps. msg. to pkts. and push to TL). Receiver TL is also Consumer (receives pkts.) and Producer (decaps. msg and delivers to AL). 2 BUFFERS, one at sending TL and one at receiving TL| Buffers ctrl. flow. If buffer full/close to OF: slow down/stop sending
msgs. If buffer empty/close to starvation: speed up/ can send pkts. again| Flow ctrl. found in 2 places: 1) between sender TCP and sender AL 2) between sender TCP and receiver TCP| Echange Data Process: TX TCP direct data to TX Buffer. Then append header to data = TCP
Segment. Segment send to TX NL. Segment encap. In IP datagram. RX IP gets decap. By RX NL, sent to RX TCP. RX TCP gets segment payload and places in TCP buffer| TCP max segment size (MSS) limited by LL maximum payload called MTU| IEEE 802.3 MTU = 1500 bytes, IEEE
802.11 MTU = 2304 bytes, IEEE 802.5 MTU = 4464 bytes| MSS = MTU – (Htl – Hnl)| standard header size for TCP or IP is 20 bytes| TCP segment structure: src. no., dest. no, seq. no., ack no., HLen, Flags, Rwnd, Checksum, Data Ptr., Options, Data| Hlen is count for no of 4 bit words in
header. Standard is 20 bytes, max = 60 bytes| With ISN = X, MSS = Y, seq no.: X, X+Y, X+2Y…| First byte in Payload is the seg. seq. no| ACK No. = Next seq. no. = current seq No + no. of data (bytes)
T TCP connection: 3 way handshake: Leg 1: SYN=1, ACK=0, Leg 2: SYN=1, ACK=1, Leg 3: SYN=0, ACK=1| TCP teardown: 4 way, teardown can start with C or S. If connection closed in 1 direction, it can still continue other way. C sends FIN = 1 to S, S sends ACK to confirm close, S
continues to send until all send, then sends FIN = 1 back to C, C sends ACK to confirm, timed wait, then closes, wait state is 30s,, 1 min or 2 min| T Cumulative ACK when one ACK is used for that seq. no. and the ones before it, for which the ACK was lost| TCP loss events: Timeout or
Three Dup. ACKs| For TO, when ACK is lost + timer done, then seq resent and ACK received| TDA: when a seq. no is lost, then ACK also lost(say ACK 100) so for next 3 seq. no, the ACK will be 100, so TDA fast retransmit send the pkt. with ACK 100 again| If TO < RTT: Premature TO, If
TO > RTT, Slow reaction to lost segments. TO[𝑘]= RTTa [𝑘] + 4Δa [𝑘] ,Timeout= RTT + safety margin| Let 𝛼 = 0.125, 𝛽 = 0.25 | RTTa[𝑘] = (1−𝛼)RTTa[k-1] + 𝛼RTTi[k]| Δa[𝑘] = (1−𝛽) ⋅Δa [𝑘−1] + 𝛽⋅Δ𝑖[𝑘]| Δi[k] = | RTTa[k] - RTTi[k] | Δa[𝑘] : average deviation | Δ𝑖 : instantaneous
deviation| average values of RTT and a used because instantaneous values have large variances. TCP ARQ Characteristics: pipelined, cumulative, timers. T If no congestion ctrl. then per conn. throughput will collapse, causing: data loss, retransmission, collapse of all connections| 2
approaches for congestion ctrl.: Classic appr. (uses sender perception) And network assisted appr. (network gives explicit feedback to TCP senders) aka ATM approach| Classic appr.: self clocking: TCP sender limits rates (either up or down) depending on perceived network cong.,
done by incr./decr. Cong. Window (Cwnd)| TCP perceives congestion when loss event occurs: either TO: severe, nothing delivered or TDA: Less severe, something delivered| Tahoe: Every RTT a segment is transmitted| Phases: Slow Start (SS): Cwnd is 1 MSS, doubles until initial
threshold, then enters Cong. Avoidance (CA): Cwnd + 1 every time| If TDA or TO occurs, Cwnd = 1, then SS starts and New threshold = ½ Cwnd[old], where Cwnd[old] is Cwnd before fall to 1| Reno: If TO: falls to 1, then SS, If TDA, enters Fast Recovery, Cwnd = ½ Cwnd[old] + 3
MSS, no SS, direct CA| Cwnd decrease is multiplicative| For Reno: Ravg = 1.5 (Wmin / RTT) | New approach - ECN: 2 bits in IP header signal congestion, 11 signals cong. Sent to Host B, then Host B echoes cong. Back to A by ECN Echo Flag ECE = 1, then Host A halves Cwnd and tells
Host B by TCP Header CWR = 1 | Segment Size = Min(Flow Window ctrl., cong. Ctrl. window) = min(Rwnd, Cwnd)| Fairness: UDP is unfair as no congestion ctrl| TCP is fair if assume that TCP connections all have same RTT. TCP Cong. Ctrl. is called additive increase and multiplicative
decrease or AIMD. Reno used AIMD HW3: is a technique that a client-side application can use to mitigate network delay jitter. buffering. Packet interleaving is a common AL technique used to mitigate packet loss in multimedia applications.
The main reason for streaming video files from online servers instead of downloading them is their large sizes. Interactive streamingaudio/video describes the type of multimedia applications such as Skype.
Packet Loss is one of the imporant issues that multimedia applications must address in today's Internet.
To support streaming multimedia content over the Internet, service providers often use CDN overlay architecture.
Delayed Playback is a technique that a multimedia application can use to mitigate network delay.
Header = 12 bytes
UDP Id: 16-bit no. to identify query, copied to reply message
of query.
