AutoSeC: An Integrated Middleware Framework for Dynamic Service Brokering

Qi Han and Nalini Venkatasubramanian

Distributed Systems Middleware Grouphttp://www.ics.uci.edu/~dsm

Dept. of Information and Computer ScienceUniversity of California-Irvine

•Quality of Service enhanced resource management at all levels - storage management, networks, applications, middleware

QoS Aware Information Infrastructure

QoS Enabled WideArea Network

BattlefieldVisualization

BattlePlanning

BattlefieldVisualization

Data servers

CollaborativeMultimedia Application

Collaborative task ClientsData servers

BattlePlanning

Global Information Infrastructure

Proliferation of devices System support for multitude of smart devices that

attach and detach from a distribution infrastructure

produce large volume of information at a high rate limited by communication and power constraints

Require a customizable global networking backbone.. Applications (e.g. multimedia) may have QoS

requirements should be translated to system level resource requirements

Explore effective middleware infrastructures which can be used to support efficient QoS-based resource provisioning algorithms

QoS-based Resource Provisioning Issues

Degree of network awareness that middleware and applications must have to deal with network conditions

Resource provisioning algorithms utilize current system resource availability information to ensure that applications meet their QoS requirements

Additional Challenges In highly dynamic (e.g. mobile) environments, system

conditions are constantly changing Maintaining accurate and current system information is

important to efficient execution of resource provisioning algorithms

Automatic Service Composition (AutoSeC) Tools needed to securely and dynamically

manage an adaptable network infrastructure while ensuring user QoS a set of network management middleware

services is critical to providing these tools AutoSeC:

dynamically select an appropriate combination of information collection and resource provisioning policies based on current system status

AutoSeC Framework

Network and Server Information Collection Policies System Snapshot (SS)

information about the residual capacity of network nodes and server nodes is based on an absolute value obtained from a periodic snapshot

Static Interval (SI) residual capacity information is maintained using a static

range-based representation Throttle (TR)

the directory holds a range-based representation of the monitored parameter, with upper and lower bounds that can vary dynamically

Time Series (MA) time series models are used to predict future trends in

sample values with some defined level of confidence.

Resource Provisioning Policies Server Selection (SVRS): attempt to choose the

best replica and server for a given request Least Utilization Factor Policy (SVRS-UF): This policy

chooses the server with the minimal utilization factor Shortest Hop Policy (SVRS-HOP): This policy chooses

the nearest server in terms of the number of hops. Combined Path and Server Selection (CPSS)

Given a client request with QoS requirements, we select the server and links that maximize the overall use of resources.

This allows load balancing not only between replicated servers, but also among network links to maximize the request success ratio and system throughput.

Performance Evaluation Objective:

To determine the best combination of information collection policies and resource provisioning policies under varying application workload

All-req-monitored: all the applications have QoS requirements Not-all-req-monitored: some requests don’t have QoS

requirements Metrics:

Request success ratio ratio of number of successful requests over the number of

whole requests Information collection overhead:

sampling overhead and directory service update overhead Overall performance efficiency:

ratio of the number of successful request to the information collection overhead

Simulation Environment Simulation topology

18 replicated data servers and 30 backbone links

Capacities of network links from 1.5Mbps to 155Mbps (mean= 64Mbps)

Capacities of servers based on real ISP data-center settings

Request and traffic generation model Request arrival as

Impact of Information Collection on CPSS Compare the performance of the four

information collection policies with the CPSS algorithm under similar conditions

All-req-monitored: Snapshot based approach is very sensitive

to sampling period Given the same sampling period, throttle

based approach is superior to other three approaches in terms of performance efficiency

Not-all-req-monitored: Exhibits similar results to above case

CPSS, All-Req-Monitored

CPSS, Not-All-Req-Monitored

Impact of Information Collection on Server Selection All-req-monitored

The overall performance efficiency of the throttle-based approach is higher than that of MA based one

Static interval based algorithm results in higher request success ratio and overall efficiency than the other three approaches

Not-all-req-monitored With fewer requests: the static interval based

approach yields higher request success ratios and performance efficiency

When more requests arrive, the request success ratio decreases and gets closer to the dynamic range based approaches

In terms of overall performance efficiency, the throttle based algorithm is better than other approaches

Impact of Information Collection on Server Selection All-req-monitored

For both svrs-hop and svrs-uf, throttle-based and MA model based approaches have similar request success ratios, but the overall performance efficiency of the throttle-based approach is higher

Static interval based algorithm results in higher request success ratio and overall efficiency than other three approaches

Only server resource factors are considered in server selection and also all requests are reflected in resource provisioning module, representing residual link bandwidth with a static interval is accurate enough

Not-all-req-monitored With fewer requests, the static interval based approach yields higher

request success ratios and also higher performance efficiency than other other dynamic ranged based approaches; but when more requests arrive, the request success ratio decreases and gets closer to the dynamic range based approaches

With a larger number of request, the success ratio is more sensitive to the application workload change.

In terms of overall performance efficiency, the throttle based algorithm is better than other approaches

SVRS-HOP, All-Req-Monitored

SVRS-UF, All-Req-Monitored

SVRS-HOPNot-All-Req-Monitored

SVRS-UFNot-All-Req-Monitored

Performance Summary Both the accuracy and overhead of information

collection policies have a significant impact on the performance of resource provisioning process

Although Snapshot based collection can obtain very accurate information, the huge overhead introduced by frequent sampling and directory updates makes it a bad choice

MA based collection does not always perform very well practically, while throttle based algorithm adapts pretty well to the constantly changing environment and turns out to be a very good choice in most cases

Optimal Combinations of Information Collection and Resource Provisioning Policies

All-req-monitored

Not-all-req-monitored

SVRS-HOP Static Interval

Throttle

SVRS-UF Static Interval

Throttle

CPSS Throttle Throttle

Preliminary Dynamic Service Composition Rules

Server type

Request type Web server Multimedia server

Computation server

General purpose server

Web request N/A N/A Multimedia

request CPSS+TR CPSS+TR CPSS+TR

Computation request

SVRS+SI SVRS+TR

Future Work

To integrate policies for AutoSeC into CompOSE|Q

To study network management middleware services applicable to mobile environment mobility management adaptive probing architecture distributed directory service management