the atha environment: experience with a user friendly environment for opportunistic computing

The ATHA Environment: Experience with a User Friendly Environment for

Opportunistic Computing

M.A.R.Dantas Department of Informatics (INE) University of Santa Catarina (UFSC)

88040-900 – Florianopolis – Brazil

Email : [email protected]

A. Hosken Department of Computer Science (CIC) University of Brasilia (UnB)

Brasilia – Brazil

Agenda

I. RESEARCH MOTIVATION

II. OPPORTUNISTIC COMPUTING

III. The ATHA ENVIRONMENT

IV. EXPERIMENTAL RESULTS

V. CONCLUSIONS AND FUTURE WORK


• It is observed that the number of computers connected to the Internet is growing for a long period of time ;

•Researches mentions that it is difficult to estimate when this trend will reverse;

•Similar to the Internet, the same phenomenon is verified inside several organisations;

•Many researchers investigate the possibility of using a large amount of resources as a useful metacomputer.


•Several distributed packages have demonstrated the efficiency to execute complex problems, even considering the ordinary Internet as a distributed computing configuration;

•It is interesting to observe that some aspects of the computational environment, such as latency of the network and co-execution of processes, may preclude the use of opportunistic computing for certain applications.

The opportunistic computing approach considers the use of spare computational resources to execute applications which could not be solved locally.

The main goal of this paradigm is to gather available heterogeneous distributed resources in computer network environments to execute distributed and/or parallel applications.


In this talk, we present experience with a user friendly environment for opportunistic computing for parallel and distributed applications in a computer network. The configuration is called as ATHA.


Agenda






II - OPPORTUNISTIC COMPUTING

In the literature it is possible to find several research projects, where the main focus is the efficient utilization of spare loosely couple resources. These projects different from a cluster (or grid) approach target to gather available disposable computing resources from users (i.e. other users and organizations) free of charge.

In other words, when we consider the use of a cluster (or a grid) configuration we understand that we pay for it.

II - OPPORTUNISTIC COMPUTING

In the opportunistic approach we plan to execute our applications considering available spare resources from other users.

This approach can be interesting even inside an organization, using the private Internet of the institution. In addition to that, it is important to mention that some patterns are required for this utilization. Special parameters such as operating system, memory and disk space requirements are some metrics that an opportunistic framework needs to address.

The goal is to have enough information to form dynamic configurations to execute parallel applications.

Agenda






III - The ATHA ENVIRONMENT

The ATHA environment for opportunistic computing is an enhanced configuration based in our previous research work.

In the configuration we use Java language, because of the portability and it allows some interesting classes (e.g. JCE, Java cryptography extension) that help with digital signature and cryptography.

Workers

mission

1

State1 State2

state2

state 1

tasks

taska resultloadWorkerInfo

tasks

Mainframe

CraySupercomputer

Laptop

Workstation

IBMCompatible

Handheld

iMac

Legion

Cluster

Globus

Condor

Master1

Master2

shadow

ATHA environment for opportunistic computing

Following our initial goal, we built an interface which could allow users to submit several services to servers, examples are:

• process new tasks,

• gather load information,

• provide information of executed applications,

• inform about the system availability.

The main ATHA interface

Agenda






In order to verify our proposed opportunistic framework, we tested the package in a large organization during a daytime period.

The target was to gather as much as possible spare machines to execute the RSAChallenge parallel application.

IV - EXPERIMENTAL RESULTS

The Distributed Environment

•Processor Architecture •IA32

•Clocks •450,700,800,1800 MHz

•Cache •512kB

•Memory •64,128, 256 MB

•Switch Throughput •10-100 MBPS

•Operating Systems •Windows 98, 2000 e XP

One feature that is present in the ATHA interface is a non intrusion characteristic for the owner of local resources.

In other words, we avoided disturbing owners of the available resources. Therefore, we start to apply this concept executing a serial Linpack benchmark to provide each machine with a weight parameter for performance.

In other words, we built a table of load indices for the machines of the environment.

Next figure shows individual performance of processors that we choose to gather resources.

Individual Performance

0

10

20 30

40

50 60

70

80 90

100

00:00 00:20 00:40 01:00 01:20 Time

MFlops 10.67.12.107 10.67.12.113 10.67.12.116 10.67.12.62 10.67.12.75 10.67.12.84 10.67.12.89 127.0.0.1

The next figure presents some details when machines enter into the pool of available resources (i.e. idle status), and when these computers become heavy loaded (i.e. busy) to execute our benchmark.

It is interesting to note that the ATHA environment considers every available machine when it was possible to provide computation aggregation.

However, when a local user starts to use their machines the processor was no longer consider into the configuration.

Global Performance (Linpack)

0

50

100

150

200

250

300

0:00 0:20 0:40 1:00 1:20 Time

10.67.12.107 busy

10.67.12.89 joint

execution

10.67.12.107

idle

10.67.12.62 busy

10.67.12.107 busy"

10.67.12.107

idle

10.67.12.62 idle

Mflops

The amount of keys that were tested are shown in the next figure, considering the use of our opportunistic approach.

This figure illustrates that we were able to test more then a million keys of the RSAChallenge parallel application. Without the ATHA environment a special (or dedicated) machine should be available to execute this application.

The environment was able to process at a peak a number of million and seven thousand keys.

% of Execution

average average

1.007.396

0

100.000

200.000

300.000

400.000 500.000

600.000

700.000

800.000

900.000

1.000.000 1.100.000

0:00 0:20 0:40 1:00 1:20 Time

Mflops

Agenda






The ATHA environment has proved to be successful executing a parallel application in a real organization and obtained interesting results.

As an improvement for the framework we are building new mechanisms to analyze more precisely workloads of the machines.

CONCLUSIONS

We are considering :

• execute several other applications from the organization;

•a version of the ATHA framework for executing in mobile devices;

FUTURE WORK

Questions ?