Constructing A Grid Simulation with Differentiated Network Service using GridSim

1. Constructing A Grid Simulation with Differentiated Network Service using GridSim Fellow of Grid Computing Grid Computing and Distributed Systems (GRIDS) Lab. The University of Melbourne, Australia Networks and Distributed Systems LabNational University of Singapore (NUS), Singapore.www.gridbus.org/gridsim/ Anthony Sulistio, Gokul Poduval, Rajkumar Buyya, Chen-Kong Tham

2. Presentation Outline Introduction Background Design and Implementation Experiments and Results Related Work Conclusion and Further Work Questions and Answers

3. Grid Information Service Grid Resource Broker Application R2 R3 R4 R5 RN Grid Resource Broker R6 R1 Resource Broker Grid Information Service database Grid as Cyberinfrastructure for e-Science and e-Business Applications

4. Resource Management and Application Scheduling • This is one of most challenging aspect of Grid Computing: • Due to presence of heterogeneity resources along dynamic variation of available capability of resources. • Application Scheduling Policies need to properly investigated/evaluated before deploying them on production Grids.

5. Varying the number of Resources (1 to 100s..1000s..). Resource capability. Cost (Access Price). Users. Deadline and Budget. Workload. Different Time (Peak and Off-Peak). We need a repeatable and controllable environment. Can this be achieved on Real Grid testbed ? Performance Evaluation: With Large Scenarios

6. Grid Environment • Dynamic: • Resource and User Properties vary with time. • Experiment cannot be repeated. • Resources are distributed and owned by different organizations. Heterogeneous users. • It is hard to create a controllable environment. • Grid testbed size is limited. • Also, creating testbed infrastructure is time consuming and expensive. • Hence, grid computing researchers turn to modeling and simulation.

7. GridSim Toolkit • GridSim is a Java-based discrete-event grid simulation package. • GridSim is based on SimJava2. • Few functionalities of GridSim: • Allows modeling of heterogeneous of various types of resources & users. • Resources can be extended to implement your own allocation policies (e.g, SLA or VO based allocation). • Supports simulation of both static & dynamic schedulers. • Simulates applications with different parallel models.

8. GridSim - System Architecture Application, User, Grid Scenario’s Input and Results Application Configuration Resource Configuration Visual Modeler Grid Scenario . . . Output Grid Resource Brokers or Schedulers’s Simulation GridSim Toolkit Application Modeling Resource Entities Information Services Job Management Resource Allocation Statistics Add your own policy for resource allocation Resource Modeling and Simulation (with Time and Space shared schedulers) Single CPU SMPs Clusters Load Pattern Network Reservation Basic Discrete Event Simulation Infrastructure SimJava Distributed SimJava Virtual Machine (Java, cJVM, RMI) Distributed Resources PCs Workstations SMPs Clusters

9. Network Functionalities • Communication networks serve as a fundamental component of grid computing. • A realistic simulation of grid environments should include the effects of sending data over shared communication lines. • Earlier versions of GridSim did not have the ability to specify a network topology, nor the functionality to connect resources through network links in the experiment.

10. Our Work • In this work, GridSim has been extended to address the above problems with the ability to simulate realistic network models by: • allowing users to create a network topology, • packetizing a data into smaller chunks for sending it over a network, • generating background traffic, and • incorporating different level of services for sending packets.

11. EA EB body() body() Send(output, data, EB) … … Receive(input, data, EA) … … Input_EA Input_EB … … … … body() body() Output_EA Output_EB body() body() (Deliver data @ t2) data, t2 Timed Event Delivery GridSim Entities Communication Model

12. New Network Extension Model • New functionalities: • support for Network Quality of Service, such as each packet has a Type of Service (ToS) attribute • support for Runtime Information, such as an ICMP ping message. • generate background traffic, which is done by Output

13. Experiment • The main aim of this experiment is to show GridSim's ability to simulate an adequate-size grid testbed. • For this experiment, we are mainly concern about the network behavior in a grid environment. Hence, we are trying to look at: • how background traffic might affect network loads and overall execution time; and • how differentiated QoS for packets might help in a heavy load situation;

14. Australian BADG test-bed – Hardware • Uni.Adelaide CS group • 2 Xeon 2.6GHz (IBM)70 GB disk • APAC/GrangeNet (at ANU) • 2 Xeon 2.6GHz (IBM)70 GB disk • Uni.Melbourne EPP group • 1 P4 Intel 2.0GHz70 GB disk • Uni.Melbourne GridBus/CS • 2 Xeon 2.6GHz (IBM)70 GB disk • Uni.Sydney HEP group • 2 Xeon 2.6GHz (IBM)70 GB disk

15. Experiment Setup • Five resources are created in four different locations: Canberra, Adelaide, Melbourne and Sydney. • All resources are connected via GrangeNet, a Gigabit wide-area network within Australia. • All links share same characteristics, i.e. MTU size of 1,500 bytes and latency of 10 milliseconds.

16. GrangeNet and Grid Modeling

17. Resource Characteristics Table 1. Australian Belle analysis data grid testbed simulated using GridSim

18. User Characteristics • There are 5 users located on each of the four locations, sharing the same characteristics: • bandwidth: 100 Mbps connected to a leaf router of each testbed site • total number of jobs: 20 each • job data size: 1 MB each • job processing power: 100 Million Instructions (MI) each • job submission: uniformly distributed among five resources as mentioned in Table 1. • background traffic: submits to all resources and other users, with inter-arrival time using a Poisson distribution approach with mean of 5 minutes. • Total number of packets for each interval is uniformly distributed in [1 ... 10]. The size of each packet is 1,500 bytes

19. Results: Advantage of network QoS in a shared network environment Table 2. Network QoS using SCFQ (self clocked fair queuing) packet scheduler (4 users out of 20 are given high priority for sending their jobs) Table 3. An Average Packet Lifetime at the Melbourne Leaf Router (which links 2 resources, hence more traffic that other leaf routers)

20. Results Table 4. An average of high priority package lifetime at the Melbourne Leaf Router under a heavier load (job data size = 10MB, previously 1 MB)

21. Results: effect of background traffic Number of packets passing through the Melbourne Leaf Router

22. Related Work Table 5. Listing of network functionalities and features for each grid simulator

23. Conclusion • GridSim toolkit provides comprehensive support application scheduling simulations in Grid computing environment. • GridSim has new features such as generating background traffic during an experiment, requesting network information during runtime and providing differentiated service for packets based on users‘ Quality of Service (QoS) requirements. • Our experiment has shown how GridSim can be used to simulate a medium-sized grid testbed. • GridSim is available to download: www.gridbus.org/gridsim/

24. Future Work • We are planning to incorporate additional features into GridSim, such as • having different types of routing algorithms, schedulers and reservation of network resources. • adding other type of network building blocks like switches and domain gateways. • support will be added for non work-conserving routers. • planning an ability to design the network topology using scripts similar to ns-2.

25. Selected GridSim Users