The Effect of Layer-2 Switches on Pathchar-like Tools Ravi S. - - PowerPoint PPT Presentation

The Effect of Layer-2 Switches

• n Pathchar-like Tools

Ravi S. Prasad,

• Constantinos Dovrolis,
• and

Bruce A. Mah

• Networking and Telecommunications Group,

College of Computing, Georgia Tech.

Packet Design Inc.

Ravi S. Prasad - ravi@cc.gatech.edu, IMW, Nov 2002 1 of 8

Overview of pathchar-like tools

• pathchar, pchar, clink - Objective : estimate link capacities
• They all use the Variable Packet Size (VPS) probing methodology
• Estimate Round Trip Time (RTT) up to hop

τ1

r

τ1

f

L ICMP Cr

1

/ dr

1

L ICMP

ICMP

d2

f

Cr

1

τr

2

τf

2

L/C2 L/C1 C1 df

1

Cr

I

C

2

==0 ==0

SND

ICMP TTL−− TTL−−

I Hop

th

• Propagation delays

• Queuing delays

• Serialization delays

Ravi S. Prasad - ravi@cc.gatech.edu, IMW, Nov 2002 2 of 8

VPS methodology

• Assume that minimum RTT for each packet size does

include queuing delays

200 400 600 800 1000 1200 1400 1600

Probing packet size (Bytes)

0.25 0.375 0.5 0.625 0.75

RTT (msec)

RTT Minimum RTT Linear fit

• Minimum RTT for packet size

up to hop

• Estimated capacity of hop

Ravi S. Prasad - ravi@cc.gatech.edu, IMW, Nov 2002 3 of 8

A single-hop path

A B

• ✁

FastEthernet Switch

Tool Capacity estimate pathchar 49.0

1.5Mbps clink 47.5

1.0Mbps pchar 47.0

1.0Mbps

• A and B have Fast Ethernet network interfaces
• Correct capacity

97Mbps (IP layer)

• All these tools consistently underestimate the link capacity
• What went wrong?
• The tool, or the estimation methodology?

Ravi S. Prasad - ravi@cc.gatech.edu, IMW, Nov 2002 4 of 8

Effect of L2 store-and-forward switches

1 1 1 1 1 α

2

α

1

L2 1,1

C

L2 1,2

C

L2 1,1

C L / L /

L2 1,2

C

L2 1,1

C L / L /

L2 1,2

C L

• ✁

Probing packet size Minimum RTT SND RCV Layer 2 Device Segment 1,1 Segment 1,2

Estimated capacity

Correct capacity

Ravi S. Prasad - ravi@cc.gatech.edu, IMW, Nov 2002 5 of 8

Experimental results at a UDel campus network path

A B C D E F :Fast Ethernet Switch; :Ethernet Switch

L3 hop Capacity L2 segments pathchar clink pchar A to B 100Mbps 5 Fast-Ethernet 17.0

• 0.0

17.0

• 0.0

17.0

• 0.4

B to C 100Mbps 2 Fast-Ethernet 62.2

• 7.2

64.7

• 9.3

62.3

• 9.1

C to D 100Mbps 1 Fast-Ethernet 100.5

• 15.0

100.3

• 22.0

101.9

• 26.0

D to E 10Mbps 2 Ethernet 5.75

• 0.15

5.6

• 0.1

5.7

• 0.1

E to F 10Mbps ?? 4.5

• 0.1

3.7

• 0.1

6.5

• 0.6

Ravi S. Prasad - ravi@cc.gatech.edu, IMW, Nov 2002 6 of 8

Other sources of error in VPS tools

• Non-zero queuing delays
• Limited clock resolution
• Error propagation/amplification from the previous hop
• ICMP generation latency
• For a detailed analysis, see Infocom 2003, San Francisco.

Ravi S. Prasad - ravi@cc.gatech.edu, IMW, Nov 2002 7 of 8

Thank you!

Ravi S. Prasad - ravi@cc.gatech.edu, IMW, Nov 2002 8 of 8