TCP Round-trip Times (RTTs RTTs) ) TCP Round-trip Times ( Popular - - PowerPoint PPT Presentation

1 1

Variability in TCP Round-trip Times Variability in TCP Round-trip Times

Jay Jay Aikat Aikat Jasleen Kaur Jasleen Kaur Don Smith Don Smith Kevin Kevin Jeffay Jeffay

Department of Computer Science Department of Computer Science University of North Carolina at Chapel Hill University of North Carolina at Chapel Hill The The UNIVERSITY UNIVERSITY of

• f NORTH CAROLINA

NORTH CAROLINA at at CHAPEL HILL CHAPEL HILL

http://www. http://www.cs cs. .unc unc. .edu edu/~ /~jasleen jasleen/research/RTT /research/RTT

2 2

TCP Round-trip Times ( TCP Round-trip Times (RTTs RTTs) )

Popular belief: Popular belief: RTTs RTTs do not vary significantly within TCP connections do not vary significantly within TCP connections

− − Mean RTT can be used to approximate per-segment

Mean RTT can be used to approximate per-segment RTTs RTTs

• TCP throughput models

TCP throughput models

[Altman00, Kumar98, Lakshman97, Mathis97, Padhye98, [Altman00, Kumar98, Lakshman97, Mathis97, Padhye98, … …] ]

• Analysis based on TCP behavior

Analysis based on TCP behavior

[Zhang02] [Zhang02]

− − RTT of initial segments is

RTT of initial segments is “ “typical typical” ” for a TCP connection for a TCP connection

[Jiang02] [Jiang02]

Do per-segment RTTs vary significantly within a TCP connection? Do per-segment Do per-segment RTTs RTTs vary significantly within a TCP connection? vary significantly within a TCP connection?

3 3

Data Source Data Source

UNC Campus UNC Campus Rest of the world

Monitor ISP router

Rich data source! Rich data source! Rich data source!

DAT ACK

4 4

Data Source Data Source

UNC Campus UNC Campus Rest of the world

Monitor ISP router

Rich data source! Rich data source! Rich data source!

DAT ACK

UNC UNC

Regional health-care center Regional health-care center Departments Research labs Departments Research labs Residence halls Residence halls Wireless users Wireless users VPNs Modem VPNs Modem ibiblio web servers ibiblio web servers

5 5

Extracting Valid RTT Samples Extracting Valid RTT Samples

• Guiding principle:

Guiding principle:

− − Consider only those

Consider only those RTTs RTTs where there is where there is unambiguous correspondence between an unambiguous correspondence between an ACK and the DAT that triggered it. ACK and the DAT that triggered it.

• Caveat: delayed

Caveat: delayed ACKs ACKs

− − Could add 200

Could add 200 – – 500 ms to RTT estimates 500 ms to RTT estimates ack [i] d a t [ i + 1 ] d a t [ i + 1 ] ack [i+2]

Monitor Remote end-point

d a t [ i + 2 ] d a t [ i ]

? ?

6 6

Trace Statistics Trace Statistics

464 M 464 M 581 GB 581 GB 236 million 236 million 258 K 258 K 1.1 million 1.1 million Connections Connections with at least with at least 10 samples 10 samples 511 M 511 M 628 GB 628 GB 252 million 252 million 962 K 962 K 22.7 million 22.7 million All All connections connections Packets Packets Bytes Bytes RTT samples RTT samples Remote Remote hosts hosts Connections Connections

Large data set! Large data set! Large data set!

7 7

Variability Across Connections Variability Across Connections

• 60% connections see min RTT less than 100 ms

60% connections see min RTT less than 100 ms

− − Only 23% see max RTT less than 100 ms

Only 23% see max RTT less than 100 ms

• ACKs

ACKs can arrive more than 25 s after DAT transmission! can arrive more than 25 s after DAT transmission!

• Mean and median

Mean and median RTTs RTTs are comparable measures are comparable measures

8 8

Variability Within Connections Variability Within Connections

• Median RTT:

Median RTT:

− − 30% of connections see a median RTT more than twice the min RTT

30% of connections see a median RTT more than twice the min RTT

• 90% RTT:

90% RTT:

− − 22% of connections see a 90% RTT more than 5 times the min RTT

22% of connections see a 90% RTT more than 5 times the min RTT

− − 90% RTT increases with min RTT

90% RTT increases with min RTT

9 9

The SYN The SYN/

/(SYN+ACK) RTT

(SYN+ACK) RTT

• RTT yielded by the SYN and SYN+ACK pair

RTT yielded by the SYN and SYN+ACK pair

− − Differs by more than 10% from min RTT for 14% of connections

Differs by more than 10% from min RTT for 14% of connections

− − Differs by more than 10% from median RTT for 50% of connections

Differs by more than 10% from median RTT for 50% of connections

10 10

Per-Segment Per-Segment RTTs RTTs: Mean or Distributions? : Mean or Distributions?

• Is mean RTT a good approximation for per-segment

Is mean RTT a good approximation for per-segment RTTs RTTs? ?

− − TCP analytical models

TCP analytical models

− − TCP evaluation (simulations)

TCP evaluation (simulations)

11 11

Ongoing Work Ongoing Work

• Impact of RTT variability on past work

Impact of RTT variability on past work − − TCP analytical models

TCP analytical models

− − Delay-based congestion control

Delay-based congestion control

− − TCP evaluation (simulations)

TCP evaluation (simulations)

− − TCP-based analysis

TCP-based analysis

• Causes of variability

Causes of variability − − Congestion?

Congestion?

− − End-hosts?

End-hosts?

• Models for per-connection

Models for per-connection RTTs RTTs − − Accurate simulation environments

Accurate simulation environments

12 12

Variability within connections Variability within connections

Standard deviation in per-connection Standard deviation in per-connection RTTs RTTs

− − Increases rapidly in the range:

Increases rapidly in the range: min RTT = 100 ms min RTT = 100 ms – – 1 s 1 s

− − Increases less rapidly in other regions

Increases less rapidly in other regions Inter-quartile range Inter-quartile range

− − Increases consistently with

Increases consistently with (med-min) RTT (med-min) RTT

13 13

Per-Segment Per-Segment RTTs RTTs: Mean or Distributions? : Mean or Distributions?

• Is mean RTT a good approximation for

Is mean RTT a good approximation for per-segment RTT? per-segment RTT?

− − TCP analytical models

TCP analytical models

− − TCP evaluation (simulations)

TCP evaluation (simulations)