AMIS: Software-Defined Privacy- Preserving Flow Measurement Instrument and Services Yan Luo, Univ. of Massachusetts Lowell Co-PI: Cody Bumgardner, Univ.

AMIS: Software-Defined Privacy-Preserving Flow Measurement

Instrument and Services

AMIS: Software-Defined Privacy-Preserving Flow Measurement

Instrument and Services

Yan Luo, Univ. of Massachusetts Lowell Co-PI: Cody Bumgardner, Univ. of Kentucky

Co-PI: Gabriel Ghinita, Univ. of Massachusetts Boston

Co-PI: Michael McGarry, Univ. of Texas El Paso

Yan Luo, Univ. of Massachusetts Lowell Co-PI: Cody Bumgardner, Univ. of Kentucky

Co-PI: Gabriel Ghinita, Univ. of Massachusetts Boston

Co-PI: Michael McGarry, Univ. of Texas El Paso

AMIS Project Objectives• 40+Gbps flow-granularity network measurement

instrument– 216-core network processor + multicore x86

• Software defined measurement APIs & libraries– Flexible specification of measurement targets and

metrics

• Preserving privacy of network flow data– Workload-aware privacy protections

• In-depth flow analytics– Project/AS utilization, patterns and trends

2

The Box

3

Software Defined Measurement

4

Privacy Protection vs. Computational Complexity

5

Strongest ProtectionSlow PerformanceLimited Query CapabilitiesOffline Mode Only

Strong ProtectionModerate PerformanceModerate Query CapabilitiesLimited Operational Mode

Best-effort ProtectionFast PerformanceFlexible Query CapabilitiesSupports Operational Mode

Perfo

rman

cePri

vacy

Differential Privacy

Searchable Encryption

Syntactic Privacy (k-anonymity, l-diversity)

• Tradeoff in privacy and computation overheads

Preserving Data Privacy in AMIS

6

Measurement Data Management and Processing

• Decentralized hierarchical resource management

• common data integration schema to be used across analytic, communication, and storage components

• Distributed system for high volume streaming data processing

• GUI, reporting tools, data repo

7

Network Flow Analytics

• Distinguish traffic matrix and traffic types– identify project association of traffic flows– file transfers, interactive sessions, short-lived

• Stochastic modeling– descriptive statistics– auto-correlation

• Derive network activity patterns and trends• Provide insights to network management

– What if scenarios

8

Test and Validation Plan

9

NSF I-Corps Interviews!

• Can I (and my student) talk with you for about 15 minutes?

10