SILO: A novel framework for flexible protocol composition

Ilia Baldine, Renaissance Computing

Institute

The SILO project isfunded by theNational Science FoundationFIND Grant

George Rouskas, Rudra Dutta,

Anjing Wang, Mohan IyerNorth Carolina State

University,

Dan StevensonRTI

Motivation• Observations of the current state

– Balkanization of protocols (VoIP, sensor nets, wireless, high-speed)

– Proliferation of half-layer solutions (IPSec, MPLS etc)

– Ossification of protocol research (e.g. growing number of TCP variants)

– Lack of explicit cross-layer interactions– Inability to separate salient features of protocols

• Looking forward– Provide ways for smooth evolution of protocols for

new transport technologies, control algorithms

Musings• Layering is important, however rigid layer

boundaries are limiting• Design for change: create a meta-design

that can accommodate future changes• Architecture must constrain – otherwise you

get proliferation of functions at various places– But must not constrain wrong thing and stifle

innovation, ossify, etc.– Job of architecture – constrain (channel) the

mechanisms used for different players to bring their various pieces and use them

Traditional and SILO network stacks

Transport

Network

Data Link

Physical

Physical Channels

App App App

Transport

Cross-ServiceTuning

silo &servicemgmt

Tuningstrategies,

hints

ComposabilityConstraints

App App App

Physical Channels

m11 m11 m13

m21 m21

m31

m21 m31 m31

m42 m43

m62 m62 m61

SILO Features• Emphasis on re-usability, not

miniaturization• Automated construction of silos via

ontology• Explicit cross-layer control and interactions• Separation of mechanisms from policies• Do not require specific OS architecture –

just a convention for implementation (template or pattern)

SILO architecture high-level view

Method

Service

S1

S2

S3

S4

S5

M1,1

M1,2 ControlAgent

Policies andStrategies

Application

M1,1

M2,2

M5,3

M3.2

M4,4

M5,1

M1,2

M2,3

M7.3

Physical Interfaces

Precedence Constraint

Service/Method Paradigm

Preceding Service Communication Spec

Service-specific Control

Interface

Succeeding Service Communication Spec

Service

MethodService-Specific Controls

Preceding Service Communications

Method Implementation

Method

Method-specific Controls Succeeding Service

Communications

The SILO hourglass• What is the convergence

point?– Service API– Ontology schema (a

language to describe network services)

10

SILO Ontology• Encodes knowledge on

relations between services and methods

• Encodes service types and functions to enable ‘fuzzy’ inference

• Written using Protégé in RDF

• Work in progress

SILO Ontology tools

SILO as a research tool• SILO is deployed with ontology and existing set of

services• Researcher brings

– Custom services– Tuning algorithms– Ontology updates

• Can connect to measurement functions to provide a cross-layer protocol experimentation tool

• Allows experimentation with various layering approaches exploring the optimization space

Wireless example: adaptive transport

• Goal: design a set of services and cross-layer tuning algorithm to maximize goodput across a wireless network– Adaptive FEC service – Adaptive MTU service – Adaptive window management service– Tuning algorithm to manage the knobs

Optical example: impairment-aware routing

• Goal: design an optical-impairment-aware routing protocol that maximizes network utilization– Distributed impairment measurement

capabilities (e.g. PMD)– Routing service capable of using the

information

Silo proof-of-concept• User-space open framework

– C++ and Python• Methods are DLLs with a well-defined interface• Silo construction agent uses ontology to create an

XML “recipe” for a silo• Silo Management Agent

– Loads and executes the code for silos based on packet events and timers

– Maintains silo state• Silo Tuning Agent is a container for tuning

algorithms/strategies associated with services• Universe of Services contains the ontology and

dynamically loadable code implementing methods

Available APIs• Application API

– Socket like for managing data transfers– Ability to specify constraints on silo construction

imposed by the application writer• Service API

– Well-defined set of functions for managing service interfaces (upper, lower, tuning knobs and gauges)

– SBuf (mbuf or skbuf like) state– Service state

SILO Prototype

Demo setup

DEMO

Future steps• Joint projects

– Converged Services Platform– Optical IMF for GENI

• Silo similarity/synchronization• Construction of siloplexes• Software-defined optics• Network Virtualization• Consistency• Stability• New services• New transports• New addressing schemes• New routing implementations• ….