Design and Evaluation in Mobility First MobilityFirst team

May 2014 FIA meeting

Presented by: Jim Kurose

Department of Computer Science

University of Massachusetts

Amherst MA USA

‘

Overview

evaluation: architecture, system, prototype

MF evaluation

GNS

control overhead, workloads

prototype demos (GENI: end-end, components)

other activities

Architecture, System, Prototype

architecture

system

prototype (realization)

high-level design/structuring principles, service/function

modularity

instantiated set of interoperating protocols, mechanisms,

platforms conforming to design principles

Implemented (sub)set of protocols, platforms in particular

existing technologies

guides, informs, inspires, constrains

“here’s what it does (function), you

tell me how”

*

* ack: J. Wroclawksi, D. Clark

Architecture, System, Prototype

architecture

system

prototype (realization)

high-level design/structuring principles, service/function

modularity

instantiated set of interoperating protocols, mechanisms,

platforms conforming to design principles

Implemented (sub)set of protocols, platforms in particular

existing technologies

*

Red team evaluation:

EAB: Crowcroft, Kobayashi, Rexford

Security red team: DeCleene, Perrig, Tsudik,

measurement, modeling:

mobility models

components: GNS, location management

implementation, demonstrations, applications

GENI, GECs

NE’s: WHYY, 5Nines, CASA

Overview

architecture, system, prototype

MF evaluation

GNS: DMap, Auspice

workloads, control overhead

prototype demos (GENI: end-end, components)

other activities

reflection: broader context

Characterizing mobility among networks

“not your father’s mobility:” characterize mobility among networks

• distinctly different from physical mobility, models

• physically mobile users may be stationary (from network transition POV); stationary users may move among networks (multi-homing, multiple devices)

two complementary activities:

• server-based: IMAP logs (across all devices): 8K users

• client-based: NomadLog Android app: 300 users

the Dave Oran question

Server (IMAP)-measured mobility

measure mobility among networks via IMAP logs

• online users periodically “push” (background login, check) email, and/or intentionally read mail

• 7100 users: two data sets (campus, SCS)

observations:

• users spend most of time in 3 networks

• surprisingly high degree of “multihoming”

Markov chain model(s) of canonical user

S. Yang, J. Kurose, A. Venkataramani, S. Heimlicher, “User Transitioning Among Networks - a Measurement and Modeling Study”, submitted to 2014 ACM IMC.

Server (IMAP)-measured mobility

Daily fraction of residence time

Network changes/ hour

Fraction of multisession time to online time

Client(NomadLog)-measured mobility

connectivity event: wakeup, connect

client connects to server, logs IP address

288 users, 10 months

Characterizing mobility among networks

“not your father’s mobility:” characterize mobility among networks

two complementary activities:

• server-based: IMAP logs (across all devices): 8K users

• client-based: NomadLog Android app: 300 users

the Dave Oran question: future network topology, ISP organization/footprint

• crystal ball: business economic, regulatory factors (discussions with Bill Lehr)

• possible approach: small number of representative cases “on the spectrum”?

Impact of Mobility on Future Architectures

use of measurement/models to assess mobility, location aspects of systems

three approaches for location-independent communication

Z. Gao, A. Venkataramani, J. Kurose, S. Heimlicher, “Towards a Quantitative Comparison of the Cost-Benefit Trade-offs of Location-Independent Network Architectures,” to appear in 2014 ACM Sigcomm

• name-based routing: device, content mobility route update costs

• name resolution: GNS update workload

• indirection routing: path stretch

A

A

Mobility event

R

B

?

B

A

A HA

FA 1 2

3 4

Indirection B

A

A

DNS

1

2

3

4

Name-to-address resolution B

A

A

1

2 3

Pure name-based routing

update cost impacted by: mobility, topology, routing preferences

approach: use RouteViews RIB data to construct FIB at dozen routers: assess % fib entry changed on mobility event

Name-based routing: device mobility update costs

content mobility: changes in IP addresses associated with name(s) for 500 most/least popular top 1M sites (Alexa)

update cost impacted by: mobility, topology, routing preferences

approach: use RIB to construct FIB at dozen routers: assess % fib entry changed on mobility event

Name-based routing: content mobility update costs

Characterizing mobility among networks

“not your father’s mobility:” characterize mobility among networks

two complementary activities:

• server-based: IMAP logs (across all devices): 8K users

• client-based: NomadLog Android app: 300 users

the Dave Oran question: future network topology, ISP organization/footprint

• crystal ball: business economic, regulatory factors (discussions with Bill Lehr)

• possible approach: small number of representative cases “on the spectrum”?

Overview

architecture, system, prototype

MF evaluation

• GNS: DMap, Auspice

• control overhead, workloads

• prototype demos (GENI: end-end, components)

• other activities

reflection: broader context

Host Stack and Network Services API

Linux PC/laptop with WiMAX & WiFi

Android phone with WiMAX & WiFi

Device: HTC Evo 4G, Android v2.3 (rooted), NDK (C++ dev)

Network API (C, Java)

E2E Transport

GUID Service

Routing

‘Hop’ Link Transport

Interface Manager

WiFi WiMAX

App-1

Security Network Layer

Device Policy + App Choice

Socket API open send recv get/put attach/ detach close

Ethernet

App-2 App-k

MF Routing and Naming Services deployed at 5 GENI rack sites with Internet2’s AL2S providing cross-site layer-2 connectivity

Wisconsin GENI rack

Utah GENI rack

BBN GENI rack

GENI Internet2 Core

GENI Edge

GENI Edge

WiMAX BTS

WiMAX BTS

MobilityFirst Software Router with GNRS instance

Dual interface Android phone with WiFi/WiMAX with MF protocol stack

ORBIT radio node with WiFi as MF Access point

Geo-Messaging Application with MF Core Services GEC-18 Oct., 2013

Controller manages local discovery of hosts, storage and compute resources

Controller presents a virtual router view for inter-domain protocols

Also interfaces with GNRS, to register local objects and lookup inter-domain bindings

OpenFlow-based prototype with

Floodlight controller • Current implementation to support

OF v1.3 with OpenDaylight controller

Performance: Comparison of forwarding ops: GUID, NA , late-binding (GNRS). Severe penalty for frequent controller interaction

SDN Design & OpenFlow Prototype

virtual router

4G/WiMAX BTS

In-network storage for store/forward

Local compute cloud

Global Name Services

MF SDN Controller

Additional prototyping

GNS implementations: DMap, Auspice

FPGA-based forwarding engine

Additional mobility-related MF activities topology models: embedding topology in physical

space

analyzing networks of caches

• approximate, bounding approaches

• populating edge caches: spatial locality

Summary

architecture

system

prototype (realization)

high-level design/structuring principles, service/function modularity

instantiated set of interoperating protocols, mechanisms, platforms conforming to design principles

Implemented (sub)set of protocols, platforms in particular existing technologies

Red team evaluation:

EAB: Crowcroft, Kobayashi, Rexford

Security red team: DeCleene, Perrig, Tsudik,

measurement, modeling:

mobility models

components: GNS, location management

implementation, demonstrations, applications

GENI, GECs

NE’s: WHYY, 5Nines, CASA

BACKUP SLIDES

Reflection: broader context

architecture

system

Internet telephony FIA

end-end circuit,

stateless endpoints,

stateful core, QoS,

single service

SS7, ESS, MSC,

VLR, HLD

datagram, stateful

endpoints, best-effort

stateless core,

multiple services

TCP, UDP, DNS,

BGP, IS-IS, OSPF

content, stateful

core, caching

Ongoing (routing, congestion

control, caching,

name resolution,

search,…)

Kleinrock 64

Cerf,&Kahn 74

Salzer 81

Clark 88, 13

McCanne 98

Kelly 98

Chiang 08

FIA Inter-architecture

comparisons under

discussion

Blocking

networks

Performance: Queueing

networks, delay calculus,

effective bandwidths, TCP,

NUM optimization

Layering: optimization

decomposition

Arch. complexity

evaluation

lots ongoing across

FIA, broader

community

variety of

goals

LOTS

Backup

architecture, system, prototype

MF evaluation

GNS: DMap, Auspice

control overhead

workloads

prototype demos (GENI: end-end, components)

other activities

reflection: broader context

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Evaluating lookup and update latency

24

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Evaluating load/capacity

25

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Architecture, System, Prototype

architecture

system

prototype

Internet telephony ICN

end-end circuit, stateless endpoints, stateful core, QoS, single service

SS7, ESS, MSC, VLR, HLR, GGSN,

many .. over the years

datagram, stateful endpoints, best-effort stateless core, multiple services, IP

TCP, UDP, DNS, BGP, IS-IS, OSPF

many .. over the years

content, naming, stateful core, caching

Ongoing (routing, congestion control, caching, name resolution, search,…)

Ramping up …

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Architecture, System, Prototype

architecture

system

Internet telephony ICN

end-end circuit, stateless endpoints, stateful core, QoS, single service

SS7, ESS, MSC, VLR, HLD

datagram, stateful endpoints, best-effort stateless core, multiple services

TCP, UDP, DNS, BGP, IS-IS, OSPF

content, stateful core, caching

Ongoing (routing, congestion control, caching, name resolution, search,…)

Kleinrock 64

Cerf,&Kahn 74 Salzer 81 Clark 88, 13 McCanne 98 Kelly 98 Chiang 08

Principles enunciated

Blocking networks

Performance: Queueing networks, delay calculus, effective bandwidths, TCP, NUM optimization

Layering: optimization decomposition

Arch. complexity Management complexity

Evaluation

ongoing, growing

simulation, prototyping, some analysis

LOTS

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Architecture, System, Prototype

architecture

system

Internet telephony MF

end-end circuit, stateless endpoints, stateful core, QoS, single service

SS7, ESS, MSC, VLR, HLD

datagram, stateful endpoints, best-effort stateless core, multiple services

TCP, UDP, DNS, BGP, IS-IS, OSPF

content, stateful core, caching

Ongoing (routing, congestion control, caching, name resolution, search,…)

Kleinrock 64

Cerf,&Kahn 74 Salzer 81 Clark 88, 13 McCanne 98 Kelly 98 Chiang 08

FIA Inter-architecture comparisons under

discussion

Blocking networks

Performance: Queueing networks, delay calculus, effective bandwidths, TCP, NUM optimization

Layering: optimization decomposition

Arch. complexity Management complexity

Evaluation

ongoing, growing

simulation, prototyping, analysis, workload analyzing GNS, control, mobility

LOTS

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

GNS Infrastructure: DMap

approach: hashing GUID to K AS-level resolvers, and local resolver for locally-created GUIDs

evaluation: performance as a function of K

simulation:

• DIMES AS-level topology, latency

• Weighted query generation, Zipf destination popularity

analysis: jellyfish topology model, latency bound

DMap: A Shared Hosting Scheme for Dynamic Identifier to Locator Mappings in the Global Internet, Vu, et al, ICDCS 2012

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

GNS Infrastructure: Auspice

approach: active name server replicas, replica controllers, consistency

evaluation: replica placement (local, spatial), #

emulation

• Planetlab, cluster

• synthetic de novo workload

• evaluate different replication approaches

• read/write latency, serverload

Sharma, Tie, Uppal, Venkataramani, A Global Name Service for a Highly Mobile Internet Venkataramani, Sharma, Tie, Uppal, Westbrook, Kurose, Raychaudhuri, Design requirements of a global name

service for a mobility-centric, trustworthy internetwork, IEEE COMSNETS 2013

101

102

103

104

100 1000 1e+04 1e+05

Load (lookup+update/second)L

ookup

late

ncy (

ms)

OptimalAuspice

CoDoNSStatic-3

Replicate-All

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Auspice: Simultaneous mobility recovery latency

31

msocket + Auspice GNS recover from simultaneous mobility in ~2 RTTs, the best achievable latency.

Serv

er

com

es

up

Clie

nt

com

es

up

Re

con

ne

ct

com

ple

te

200ms

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Evaluating Control Overhead

Question: overhead when routing on names vs. adding a topological address layer ?

Router-router name exchange:

- Hierarchical names

- Used for routing packets

- Used for caching at routers

Hybrid GUID-Name (HGN) Approach:

- Use flat GUIDs for caching

- Use topological addresses for routing

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

An Analytic Model

• Model: 𝑁levels of hierarchy in the name; prefix at level 𝑖 having 𝑙𝑖 sub-level prefixes.

• Define 𝑛𝑡𝑜𝑝 ∈ {1,2,… ,𝑁} which indicates the prefix level

below which the naming tree starts being influenced by the network topology

UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science

Sample Result

1 2 3 4 5 6 7 8 9 1010

0

105

1010

1015

1020

ntop

value

Nu

mb

er

of

En

trie

s (

log

sc

ale

)

L = 10

L = 50

L = 100

HGN (name independent)

Routing Table Size with

Topology Independent Prefixes

Current BGP

Table Size

Hierarchy in name reduces the table size only when the name prefixes

have some degree of dependence on the physical network topology.

A. Baid, T. Vu, D. Raychaudhuri, “Comparing Alternative Approaches for Networking of Named Objects in the Future Internet,” 2012 IEEE Infocom NOMEN