Strategized, Coordinated Services for Real-Time Middleware · Priyanka Gupta, Jiangyin Zhang,...

Strategized, Coordinated Services for Real-Time Middleware

Lisa Cingiser DiPippo, Victor Fay-Wolfe, Matthew Murphy, Priyanka Gupta, Jiangyin Zhang, Jianming Ye, Kevin Bryan, Tim Henry

Sponsors:

! Distributed end-to-end middleware real-time enforcement using global scheduling information

! Coordinated middleware services ! Coordinated scheduling points! Priority-based scheduling! Classic real-time analysis and

algorithms

URI Real-Time Middleware Approach

3

URI Real-Time Middleware Services

{Scheduling

Service

RT Trader(Binding)Service

Data DistributionService

URI implements coordinated distributed end-to-end real-time as middleware common object services

RT Notification Load Balancing

Object(Servant)

IDLStubs

IDLSkeletons

ORB Core

in args

out args + return value

Operation ()

BSS or Spawn

ESS

USS

Client

Service Context1

2

3

4 76 Object

Adapter

RTC2 Scheduler

Mutex

RTC2 Scheduler

NIC NIC

Router

913

14

10 12

NotificationService

16

RT Trader(Binding) Service

0

Many Scheduling Points in End-To-End Chain

Client Server (E2E-1)

5

8Object

(Servant)

IDLSkeletons

ObjectAdapter

Mutex

RTC2 Scheduler

NIC

in args


Operation ()

Service Context

Distributable Thread (E2E-N)

Object(Servant)

IDLStubs

IDLSkeletons

ORB Core

in args


Operation ()

BSS or Spawn

ESS

USS

Client

Service Context1

2

3

4 7

5

6 ObjectAdapter

8

RTC2 Scheduler (e.g. Kokyu)

Mutex


NIC NIC

Router

913

14

10 12

Object(Servant)

IDLSkeletons

ObjectAdapter

Mutex


NIC

in args


Operation ()

Service Context

Dynamic Distributed Scheduling Service

SchedulingService

Coordinated Distributed Scheduling

NotificationService

16

RT Trader(Binding) Service

0

SchedulingService

Real-Time

Sets CORBA Priority

Sets Priority Inheritance, Priority Lanes, Priority Bands

Sets OS Prio Mapping

Sets Network Prio Mapping

Offline tools set parameters for scheduling service; service sets parameters in RT ORB

URI RT CORBA 1.0 Scheduling Service

Priorities, ceilings, mappings, XML file

6

Distributed Scheduling Service:

• Set end-to-endscheduling parameters for local schedulers using global criteria

• Detect overload

• Determine distributable thread cancellation points

Real-Time

SchedulingService

Scheduling Service works with local schedulers to enforce global scheduling

Dynamic Distributed Scheduling Service

RT CORBA 2.0 Pluggable Scheduler

7

8

System

Local Only E2E-N

Static Dynamic

All Hard All Soft

All Periodic Periodic/Sporadic All Sporadic Periodicity

Remote Calls

System Conditions

Constraints

Distributed Scheduling Problem Space Taxonomy

E2E-1(per request)

Hard/Soft

Identify application and system characteristics

9

System

Local Only E2E-N

Static Dynamic

All Hard All Soft

All Periodic Periodic/Sporadic All Sporadic

RT CORBA 1.0 Scheduling Service

E2E-1(per request)

Hard/Soft

• Handles local scheduling and one CORBA call, static systems, all forms of constraints depending on the offline capability

• Example: scheduling static UAV video stream

10

System

Local Only E2E-N

Static Dynamic

All Hard All Soft


Dynamic Scheduling Service: First Problem Space Approached

E2E-1(per request)

Hard/Soft

• First dynamic problem space – dynamic periodic

•Example: dynamically entering UAVs each with periodic video stream.

Overload

• Load Shedding

•QoS adjustment

• Resource reallocation

Local Dispatch

• Priority Mapping

• RTC2 local ORB scheduler

Server Dispatch

• Priority Inheritance

• Release-Guard

• E2E intermediate deadlines

SchedulingService


Binding

• Best Fit Utilization

• Worst Fit Utilization

• Best fit laxity

Network Routing

• Net Priority Mapping

• Network EDF with intermediate deadline

System

Local Only E2E-N

Static Dynamic

All Hard All Soft


E2E-1(per request)

Hard/Soft

Distributed Scheduling Service Strategization PointsPriority

• Rate Mono

• EDF

• MUF

Overload

• Load Shedding

•QoS adjustment

• Resource reallocation

Local Dispatch

• Priority Mapping

• RTC2 local ORB scheduler

Server Dispatch

• Priority Inheritance

• Release-Guard

• E2E intermediate deadlines

SchedulingService


Binding

• Best Fit Utilization

• Worst Fit Utilization

• Best fit laxity

Network Routing

• Net Priority Mapping

• Network EDF with intermediate deadline

Distributed Scheduling Service Strategization Points

Priority

• Rate Mono

• EDF

• MUF

Current Work –December 03 target

Real-Time

SchedulingService

RT CORBA 2.0 Pluggable SchedulerRT Trader(Binding)Service

13

G lo b a lS c h e d u lin g

b e g in _ g lo b a l_ s c h e d u lin g _ s e g me n t()c a lc u la te _ u til()

S ys te mR e p o s ito ry

u p d a te _ ta s k ()b e g in _ ta s k ()g e t_ u tiliza tio n ()

O v e rlo a dD e tec to r

s ch e d ulab ility _ te st ()

O v e rlo a d H a n d le r

a c ce p t_ ta s k()d e ny _ ta sk ()

S c h e d u lin g Alg o rith m

g e t_ sc h e d_ p a ra m()

D S _ Ad a p te r

b e g in _ n e w _ s c h e d u lin g _ s e g me n t()

RT CORBA 2 Local Scheduler Interaction With Distributed Scheduling Service -

Partial Class Diagram

S c h e d u lin g A lg o rit h mO v e rlo a d H a n d le rO v e rlo a d D e t e c t o r S y s t e m R e p o s i t o ryG lo b a lS c h e d u lin gD S _ A d a p te r R T C 2 -S c h e d u le rC lie n t O R B

1 : b s s1 : b s s 1 . 1 : b n s s1 . 1 : b n s s

1 . 1 . 1 : b g s s1 . 1 . 1 : b g s s

1 . 1 . 1 . 1 : c a lc u la t e _ u t il()1 . 1 . 1 . 1 : c a lc u la t e _ u t il()

1 . 1 . 1 .1 . 1 : 1 . 1 . 1 .1 . 1 :

1 . 1 . 1 . 2 : b e g i n _ t a s k1 . 1 . 1 . 2 : b e g i n _ t a s k

1 . 1 . 1 . 3 : s c h e d u la b ilit y _ t e s t1 . 1 . 1 . 3 : s c h e d u la b ilit y _ t e s t1 . 1 .1 .3 . 1 : g e t _ u t ili za t io n1 . 1 .1 .3 . 1 : g e t _ u t ili za t io n

1 . 1 . 1 . 3 . 1 . 1 : 1 . 1 . 1 . 3 . 1 . 1 : 1 .1 .1 . 3 . 2 : 1 .1 .1 . 3 . 2 :

1 . 1 . 1 . 4 : if o v e rlo a d : d e n y _ t a s k1 . 1 . 1 . 4 : if o v e rlo a d : d e n y _ t a s k

1 . 1 . 1 . 4 . 1 : S c h e d u le _ F a ilu re E x c e p t io n1 . 1 . 1 . 4 . 1 : S c h e d u le _ F a ilu re E x c e p t io n

1 .1 .1 .5 : g e t _ s c h e d _ p a ra m1 .1 .1 .5 : g e t _ s c h e d _ p a ra m

1 . 1 .1 .5 . 1 : 1 . 1 .1 .5 . 1 :

1 . 1 . 1 . 6 : 1 . 1 . 1 . 6 : 1 . 1 . 2 : b e g in _ n e w _ s c h e d u lin g _ s e g m e n t1 . 1 . 2 : b e g in _ n e w _ s c h e d u lin g _ s e g m e n t

1 . 1 . 2 . 1 : 1 . 1 . 2 . 1 : 1 . 1 . 3 : 1 . 1 . 3 :

1 . 2 : 1 . 2 :

Scheduling Point #1 - begin_scheduling_segment

Scheduling Point #2 - update_scheduling_segment

O R BC lie n t R T C 2 -S c h e d u le rD S _ A d a p te r G lo b a lS c h e d u lin g S y s t e m R e p o s i to ryO v e rlo a d D e te c to r O v e rlo a d H a n d le r S c h e d u lin g A lg o rith m

1 : u s s1 : u s s1 . 1 : u s s1 . 1 : u s s

1 . 1 . 1 : u g s s1 . 1 . 1 : u g s s1 . 1 . 1 . 1 : c a lc u la t e _ u t il()1 . 1 . 1 . 1 : c a lc u la t e _ u t il()

1 . 1 . 1 .1 . 1 : 1 . 1 . 1 .1 . 1 :

1 . 1 . 1 . 2 : u p d a t e _ ta s k1 . 1 . 1 . 2 : u p d a t e _ ta s k

1 . 1 . 1 . 3 : s c h e d u la b ility _ te s t1 . 1 . 1 . 3 : s c h e d u la b ility _ te s t1 . 1 .1 .3 . 1 : g e t_ u t ili za t io n1 . 1 .1 .3 . 1 : g e t_ u t ili za t io n

1 . 1 . 1 . 3 . 1 . 1 : 1 . 1 . 1 . 3 . 1 . 1 : 1 .1 .1 . 3 . 2 : 1 .1 .1 . 3 . 2 :

1 . 1 . 1 . 4 : if o v e rlo a d : c a n c e l_ ta s k1 . 1 . 1 . 4 : if o v e rlo a d : c a n c e l_ ta s k

1 . 1 .1 .4 . 1 : T h re a d _ C a n c e l E xc e p t io n1 . 1 .1 .4 . 1 : T h re a d _ C a n c e l E xc e p t io n

1 .1 .1 .5 : g e t_ s c h e d _ p a ra m1 .1 .1 .5 : g e t_ s c h e d _ p a ra m

1 . 1 .1 .5 . 1 : 1 . 1 .1 .5 . 1 :

1 . 1 . 1 . 6 : 1 . 1 . 1 . 6 : 1 . 1 . 2 : u p d a te _ s c h e d u lin g _ s e g m e n t1 . 1 . 2 : u p d a te _ s c h e d u lin g _ s e g m e n t

1 . 1 . 2 . 1 : 1 . 1 . 2 . 1 : 1 . 1 . 3 : 1 . 1 . 3 :

1 . 2 : 1 . 2 :

S y s t e m R e p o s ito ryG lo b a lS c h e d u lin gD S _ A d a p te rC lie n t O R B

1 : e s s1 : e s s

1 .1 : e s s1 .1 : e s s

1 .1 . 1 : e g s s1 .1 . 1 : e g s s

1 .1 .1 .1 : re m o v e _ t a s k1 .1 .1 .1 : re m o v e _ t a s k

1 .1 . 1 .1 . 1 : 1 .1 . 1 .1 . 1 :

1 .1 . 1 .2 : 1 .1 . 1 .2 : 1 .1 . 2 : 1 .1 . 2 :

1 .2 : 1 .2 :

Scheduling Point #3 - end_scheduling_segment


SchedulingService

G lo b a lS c h e d u lin g

b e g in _ g lo b a l_ s c h e d u lin g _ s e g me n t()c a lc u la te _ u til()

S ys te mR e p o s ito ry

u p d a te _ ta s k ()b e g in _ ta s k ()g e t_ u tiliza tio n ()

O v e rlo a dD e tec to r

s ch e d ulab ility _ te st ()

O v e rlo a d H a n d le r

a c ce p t_ ta s k( )d e ny _ ta sk ()

S c h e d u lin g Alg o rith m

g e t_ sc h e d_ p a ra m()

D S _ Ad a p te r

b e g in _ n e w _ s c h e d u lin g _ s e g me n t()

System Repository

UML For Coordinated

Services

18

Real-Time Data Distribution

RTNS

EventChannel

NCA::EventChannel

NCA::ConsumerAdmin* NCA::SupplierAdmin*

ECA::EventChannel

ECA::ConsumerAdmin ECA::SupplierAdmin

ECA::ProxyPushConsumer*

ECA::ProxyPullConsumer*

NCA::ProxyPullConsumer*

NCA::ProxyPushConsumer*

ECA::ProxyPushSupplier*

ECA::ProxyPullSupplier*

NCA::ProxyPushSupplier*

NCA::ProxyPullSupplier*

EC::PushSupplier*

NC::PushSupplier*

EC::PullSupplier*

NC::PullSupplier*

EC::PushConsumer*

EC::PullConsumer*

NC::PushConsumer*

NC::PullConsumer*

Direction of Event Flow

EC - CosEvenCommECA- CosEvenChannelAdmin(Pure Event Service)NC-CosNotifyCommNCA-CosNotifyChannelAdmin*-multiple object instances supporting this interface

Supplier provides real-time data to

consumers

Consumer specifies real-time requirements for

data delivery

SchedulingService

Scheduling service determines scheduling

parameters for propagation of event

data

Model Integrated Scheduling - Approach

System

Local Only E2E-N

Static Dynamic

All Hard All Soft


E2E-1(per request)

Hard/Soft

Modeled Problem Space Taxonomy in GME2000

20

Model-Integrated Scheduling- Current

I

I

SchedulingAspectDelegates

RapidRMA Analysis

I URI model interpreterTri-Pacific model interpreter

Modified BBN QuO model interpreter

XML file

reads

XML file

reads

SchedulingService


System

Local Only E2E-N

Static Dynamic

All Hard All Soft


E2E-1(per request)

Hard/Soft

Determine Scheduling Problem Space

I

TimeWiz Analysis

reads

Model-Integrated Scheduling On- going Work

TimeWiz

U Virginia VEST/TimeWiz

Integration

VEST

(U Virginia)

I

Scheduling Aspect Delegates

I

SchedulingService


22

Summary! Creating Real-Time Services

! Scheduling, Binding, Data Distribution

! Embodying classic real-time algorithms in middleware

! Creating new algorithms as needed! Using existing services for framework

and enforcement

Strategized, Coordinated Services for Real-Time Middleware

Lisa Cingiser DiPippo, Victor Fay-Wolfe, Matthew Murphy, Priyanka Gupta, Jiangyin Zhang, Jianming Ye, Kevin Bryan, Tim Henry

Sponsors:

Contact

Vic Fay-Wolfe ([email protected])

Lisa DiPippo ([email protected])http://rtdoc.cs.uri.edu

Strategized, Coordinated Services for Real-Time Middleware · Priyanka Gupta, Jiangyin Zhang,...

Documents

Transcript of Strategized, Coordinated Services for Real-Time Middleware · Priyanka Gupta, Jiangyin Zhang,...