Introduction

We started exploring Cycle Harvesting Grid Computing since 2003. In that year, a testbed with 120 computers was set up using the Platform Computing ActiveCluster middleware. At the end of 2004, a collaboration was established between NUS and Singapore Computer Systems (SCS) to expand the Cycle Harvesting Grid (code named TCG@NUS, which stands for Tera-Scale Campus Grid at NUS) to 800 computers in Phase 1 of the project and to 3000 PCs/servers subsequently.

Cycle Harvesting Grid Computing is about harnessing the idle compute cycles of the PCs/servers to do large scale research and other intensive computations. Some of the well-known Cycle Harvesting Grid projects include SETI@home, Einstein@home and Life Sciences related projects at grid.org. In our implementation, applications from Life Sciences, Financial Engineering, Image Processing & Rendering, data-mining, statistical analysis and parametric studies in various research areas will be explored and Grid-enabled to run on the Cycle Harvesting Grid. We believe that the Cycle Harvesting Grid will provide a new computing capability that will enable NUS researchers to push the boundary of many computation based research.
 

♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦
Click here to join your computer to NUS Grid Computing
♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦♦
 

The Current TCG@NUS Setup

The Grid Middleware and Architecture

Basically there are three groups of stakeholders in this setup: the resources owners, the application owners and the users.

Resource owners: They are the computer owners such as Computer Centre, Central Library and Schools/Faculties, which have contributed computers to the Cycle Harvesting Grid. The current setup is as follow:

Faculties/Schools/Departments	No. of PCs/Servers Contributed
Biochemistry	49
Business School	104
Centre for Instructional Technology (CIT)	36
Central Library	107
Centre for English Language Communication (CELC)	20
Computer Centre	338
Department of Mathematics	97
Department of Physics	45
Faculty of Engineering	396
School of Computing	234
School of Design & Environment	77
Computer Centre Staff PCs and Others	17
Total	1520

Application owners: They are either the researchers who have developed their own application software or the resource owners such as SVU at Computer Centre who have installed applications that can be used by everybody.

Users: Staff and students of NUS. A SVU user account is needed to access the Cycle Harvesting Grid resources. Information on account application can be found at SVU webpage.

How Does It Work?

1. A UD Grid MP Platform Server software is installed on the Cycle Harvesting Grid Server. A Client Agent is installed on each of the computers/servers.
    
2. Before an application can run in parallel mode on the Cycle Harvesting Grid, it has to be Grid-enabled. The effort includes writing a script/wrapper program (Program Module Executable) to split the application into multiple tasks (workunits). Both the application and the script/wrapper program have to be installed on the Cycle Harvesting Grid Server before execution. This step will not be necessary if you are running an application that has been Grid-enabled.
    
3. When a job is submitted, the splitting of the job into multiple tasks (Workunits) will be done automatically by the script/wrapper program. The Client Agent on the computer/server will contact the server (the Dispatch/Scheduling service) to receive a Workunit to run when the load on the computer/server is below a preset level. The result will be sent back to the server upon completion.

Calling for Contribution from Resource Owners

The Cycle Harvesting Grid was initiated with the aim to create new High Performance Computing (HPC) capability for the University using the existing resources. Such high-performance computing capability is essential to NUS as it will allow researchers to push the frontier of research and development in many areas. As the new capability is created by harnessing the idle cycles of the PCs/servers, we need to connect up as many PCs/servers as possible in order for it to be useful. With the Cycle Harvesting Grid, NUS researchers will be able to access hundreds or thousands of PCs/servers for large scale parallel computation right from their desktops. The aim is to link up 3000 PCs/servers by 2007. To achieve that target, we need more computer owners in NUS to contribute their resources to the Cycle Harvesting Grid. The contribution can be in the form of departmental computer cluster or individual computer.

Minimum effort will be required from the computer owners during and after the connection to the Cycle Harvesting Grid. For departmental computer cluster, the one-time Client Agent software installation will be done by SVU. For individual computer/notebook, a simple point-and-click installation process will take less than one minute to complete. Student notebook with wireless connection can also join and contribute to the Cycle Harvesting Grid.

Performance Concern

The UD Grid MP middleware allows the systems administrator and the owners to control how and when the resources will be used. The local users will always be given the first priority. The job will stop running on a particular computer when a local user or the owner is actively using the computer. We have not received any complaint on the performance since our implementation from early 2005.

Security Concern

Security is a genuine concern in such a distributed environment. In the TCG environment, there are security measures to protect both the resource owners and the users. The following measures ensure that the computer resources are protected from any malicious software and the user data privacy is maintained:

	All application jobs are executed within a secure runtime 'sandbox' environment provided by the Grid MP Platform agent. This provides two security benefits: it protects job execution on PCs/servers from being tampered with, and it protects unshared portions of the PCs/servers from being accessed by the jobs.
	All job data are automatically encrypted on the PCs/servers, preventing access by the owner/local user.
	Grid MP Platform provides both resource load control and automatic monitoring against malicious application jobs.
	Thresholds for the processor, memory and disk usage on PCs/servers are set by the administrators so that jobs cannot hog node resources (whether or not intentional, e.g. a 'Denial of Service' attack).
	Digital signature verification of application code is carried out before execution on the PCs/servers. This ensures that only authorised and legitimate code is allowed to execute. Users can only launch jobs from the approved application code.

Reference

TCG@NUS launch: http://newshub.nus.edu.sg/ke/0406/articles/pg11.htm

HPC@NUS newsletter: http://www.nus.edu.sg/comcen/news/HPC/

PovRay performance benchmark: http://www.nus.edu.sg/comcen/svu/publications/hpc_nus/...

Modeller case study: http://www.nus.edu.sg/comcen/svu/publications/hpc_nus/...

Structured product pricing simulation: http://www.nus.edu.sg/comcen/svu/publications/hpc_nus/...

United Devices website: http://www.univaud.com/hpc/products/grid-mp/

Applications and Case Study

Visit the TCG@NUS Apps page to learn about the applications that were enabled on the Grid and also what kind of applications can be enabled to run on the Grid.

Please contact us at if you like to know more about Grid Computing.