컴퓨터 네트워크 ch1(4)

Performance : Delay, loss throughput

---

Performance in Networks

• 네트워크 성능 측정 기준
  • Throughput, delay, loss
• Throughput
  • 네트워크를 통해 실제적으로 얼마나 많은 데이터를 보낼 수 있는가
  • "bit per second"
• Delay (Latency)
  • 한 패킷을 목적지까지 보내는데 얼마나 시간이 걸리는가
  • Delay = transmission time + propagation delay + queueing delay + processing delay 

 

Four sources of packet delay

1. Processing delay

• 라우터에서 발생하는 지연
• packet이 도착하면 에러가 났는지 확인
• table을 통해 어디로 갈지 확인

2. Queueing delay

• 총 도착한 packet가 outlink capacity보다 커질 때 buffer에 담는다
• queue에서 기다리는 시간
• congestion이 발생하면 queueing delay가 생긴다
• queue 사이즈를 얼마나 크게 할 것인가가 고민
• ㎲부터 ㎳까지 생길 수 있다
  • delay variation jitter
  • 인터넷 상에서 packet이 빨리 도착했다가 늦게 도착했다 하는 것들은 queueing delay 때문

3. Transmission delay

• packet을 전송할 때 걸리는 시간
• R = link bandwidth (bps)
• L = packet length (bits)
• Transmission delay = L/R

 

4. Propagation delay

• source부터 destination까지 bit가 전달 될 때 걸리는 시간
• d = physical link의 길이 (path)
• s = 전파가 매체를 통해서 전달되는 속도 ( ~ $2\times10^8 m/s$)
• propagation delay = d/s

 

 

Delay in packet-switched networks

Example [one packet, one hop]

• 1Mbit 파일을 64kbps 링크를 통해 보낸다
• source와 destination 사이의 거리는 4800km이고
• link의 속도는 $2\times10^8m/s$
• processing delay를 무시해라

$d_{total} = d_{prop} + d{tans}$

$d_{prop} = d/s = 4800\times10^3m/2\times10^8m/s = 24[ms]$

$d_{trans} = 1\times10^6bits/64\times10^3=15.625[sec]$

$d_{prop}$ 보다  $d_{trans}$가 훨씬 크다. propgation delay는 무시할 수 있다!

 

 

Example

• under the same condicion except the link speed
• Link speed = 1Gbps

$d_{total} = d_{prop} + d_{trans}$

$d_{prop} = d/s = 4800\times10^3/2\times10^8=24[ms]$

$d_{trans} = 1\times10^6/1\times10^9=1[ms]$

 

high-speed links에서는 propagation delay를 무시할 수 없다!!!

 

 

 

Delay comparison

Circuit Establishment + packet switching으로 보내는 시간

 

 

Example [multiple hops, multiple packets]

• Number of hops = N = 3
• Message Length = L = 5500 [bits]
• Link rate = R = 9600 [bits/sec]
• Packet size (payload + header) = P = 1040 [bits]
• Header overhead = H = 40 [bists]
• Circuit establishment = $d_{setup}$ = 0.2 [sec]
• Propagation delay per hop = $d_{prop}$ = 1 [ms]
• queueing delay와 processing delay는 무시한다

 

 

$d_{total} = d_{setup} + d_{t_prop} + d_{trans}$

$d_{setup} = 0.2sec$

$d_{t_prop} = 3\times1m/s$

$d_{trans} = 5500bits/9600bps = 0.573sec$

 

$d_{total} = 0.776 [sec]$

 

 

 

40 + 1000 [ P1 ~ P5 ]

40 + 500 [ P6 ]

 $d_1 = transmission n_time_for_one_packet\times(N-1)+N\timesd_{prop}$

$d_1 = 1040/9600\times2+3\times0.001=0.220[sec]$

$d_2=1040\times5+540/9600[sec]$

 

 

 

packet switching 하는데 걸리는 시간 + call setup 시간

 

 

 

 

Queueing delay (revisited)

• Very complex component in network delay
• R = link bandwidth (bps)
• L = packet length (bits)
• a = average packet arrival rate
• traffic intensity = La/R

 

• La/R ~ 0 : average queueing delay small
• La/R -> 1 : delays become very large
• La/R > 1 : more "work" arriving than can be serviced (congestion)
• For infinite queue : delay is infinite
• Finite queue : loss is very large

 

"Real" Internet delays

• Traceroute (tracert for window)
  • Utility program that provides delay measurement from source to router along end-end Internet path towards destination
• Traceroute operation
  • sends three packets that will reach router i on path towards destination
  • router i will return packets to sender
    • When a router sees a packet with TTL=0, it discards the packet and send an ICMP packet to the source
    • 라우터를 통과할 때마다 TTL을 1만큼 감소
  • sender times interval between transmission and reply

 

 

Packet loss

• queue (buffer) has finite capacity
• when packet arrives to full queue, packet is dropped (lost)
• lost packet may be retransmitted by previous node, by source end system, or not retransmitted at all

 

Throughput

• throughput : rate (bits/time unit) at which bits transferred between sender/receiver
• instantaneous : rate at given point in time
• average : rate over long(er) period of time)

bottleneck link : link on end-end path that constrains end-end throughput

 

 

Throughput : Internet scenario

• per-connection end-end min(Rc, Rs, R/n)
• in practice : Rc or Rs is often bottleneck