Understanding Block-level Address Usage in the Visible InternetXue Cai and John Heidemann, SIGCOMM ’10

B95502017 機械五 Chen-Hsin Ding (丁振新)

Network Issues in Cloud Computing 2011

Final Presentation

Outline

Introduction

Methodology

Applications

Validation

Conclusion

Introduction (1-4)This work provides:

Little insight into the edge of the Internet

The use of the IPv4 address space

Assumptions:

1.Many active addresses will respond to probes

2.Contiguous address are often used similarly

3.Patterns of probe responses and response delay suggest address usage

Introduction (2-4)

Introduction (3-4)Approach and Validation:

Surveyed around 24,000 /24 address blocks

Pinged every 11 minutes for around one week

Applications:

To understand how addresses are managed

How effectively addresses are used

Detect and quantify the use of dynamic address assignment

Applications (cont.):

Distinguish blocks connected mainly by low-bitrate edge links from those with broadband connections

Introduction (4-4)

The contribution of this paper is therefore to develop new approaches to classify Internet address usage and to apply those approaches to answer important questions in network management.

Methodology (1-11)

Data Collection: Surveying the Internet

Representation: Observations of Intrest

Block Identification

Ping-Observable Block Classification

Identifying Low-bitrate Blocks

Ping each address of about 1% of the allocated Internet address space around every 11 minutes for one week longer

Ignore all non-positive (no reply) responses.

Most of the /24 blocks did respond at least once by one IP address

Methodology (2-11)Data Collection: Surveying the Internet

Address Usage Metrics:

1.Availability

• The fraction of time an address is responsive.

2.Volatility

• A normalized representation of how many consecutive periods the address is responsive.

3.Median-up

• The median duration of all up periods.

Methodology (3-11)Representation: Observation of Interest

Availability

Volatility

Median-up

Methodology (4-11)Representation: Observation of Interest

Availability shows how effectively address are used.

High volatility indicates addresses that are intermittently used and often dynamically allocated.

Median uptime suggests how long an address is used.

Methodology (5-11)Representation: Observation of Interest

Edge Bitrate Metrics:

1.Median-RTT

2.Stddev-RTT

Median-RTT tracks typical response bitrate

Stddev-RTT estimates variance

Methodology (6-11)Representation: Observation of Interest

The process of finding a prefix where addresses in the block are used consistently.

The algorithm is based on elbow criterion, not K-means.

Methodology (7-11)Block Identification

Methodology (8-11)Block Identification

4 thresholds:

αH = 0.95 for high availability

αL = 0.10 for low availability

β = 0.0016 for low volatility

γ = 6 hours for a relatively long uptime.

Methodology (9-11)Ping-Observable Block Classification

Categories:

Always-stable

Sometimes-stable

Intermittent

Underutilized

Unclassifiable

Methodology (10-11)Ping-Observable Block Classification

Definition:

Low bitrate: <100Kb/s

eg.: dial-up (56Kb/s) and GPRS (57.6Kb/s)

Round-trip time (RTT) = 2(Dcpu+Dprop+Dt+Dq)

Low-bitrate block:

Goal:

To distinguish addresses with low-bitrate edge links from broadband links.

Methodology (11-11)Identifying Low-bitrate Blocks

(δ=300ms)

Application

Block Sizes

Address Utilization

Intermittent and Dynamic IP Addressing

Understanding Edge Bitrates

ApplicationBlock Sizes

A lower bound of 61% of the responsive Internet is used consistently.This supports the second assumption: the majority of contiguous addresses are used consistently.

ApplicationBlock Sizes

ApplicationAddress Utilization

ApplicationAddress Utilization

ApplicationIntermittent and Dynamic IP Addressing

ApplicationUnderstanding Edge Bitrates

Conclusion

Contiguous addresses are often used similarly

This work provides a new tool to understand Internet use and trends.

Thank You!