Grid Middleware

Goal

Grid middleware aims to achieve virtualization of the software/hardware platforms so that the VO will only need a unique grid application to achieve policy enforcement in a delegation manner.

Overview

Traditional gird computing is mainly concerned about job execution and ignores deployment of the execution environment while there are many kinds of applications with different resource requirements, and each of them requires a very specific, customized environment. As variations in operating system, library environments, and file system layouts a user's application may use only a small fraction of the resources in the Grid. It also needs to provide reliable isolation and dynamic, fine-grain control of shared resources to ensure enforcement of policies. So it's necessary for the gird middleware to provide virtual execution environment based on virtual machine that can be made dynamically available to authorized clients.

Our implementation is based on the CGSP(ChinaGrid Support Platform) which is a grid middleware developed for ChinaGrid to support the construction of each campus network and grid applications in China, as depicted in Figure 1. The authorized Grid User configures the Remote virtual machine via VM Factory Service, which allows a grid client to deploy a VM-based execution environment on a suitable and trusted node according to a deployment request specifying resource allocation, environment definition and security policy. Grid Daemon installed in dom0 of backend nodes is a control agent for communicating with frontend node to perform task management including deployment and control of virtual machine. We can use VM Control Service to pause, shutdown, restart a VM, and an application submitted by the client will run on the VM. The process isolation and trusted computing technologies provided by FDU and WHU will be used to protect the privacy of user application. Any time the backend has no capacity to continually run the task, The job and it's execution environment will be migrated securely to another trusted backend node, which should be transparent to the end users.

Features

Following are the main features of virtual execution environment in CGSP:

1. Resource register and monitoring

Anyone who wants to "go for grid" to become a lessor should register computing/storage resources to grid broker, and the broker monitors the current resource status of all the backend nodes in resource pool for scheduling.

2.Remote deployment and lifecycle management of VMs

VM Factory and Control Services are some WSRF services running on the CGSP Container. These services provide the client interfaces to manage VMs. With these services, clients can deploy, pause, restart and shutdown VMs.

3. Job management

After a virtual machine is deployed, a user application will be submitted to and run on the VM. The process isolation and trusted computing technologies provided by FDU and WHU are used to protect the privacy of user application.

4. Request authentication

Grid broker is used to authenticate grid users' identities before deploy requests are accepted.

5. VM fine-grain resource allocation

Grid broker allows the client to specify the resource allocation to be assigned to a VM and manage that resource allocation during deployment.

6. Policy configuration

Policy files submitted by grid users specify the conditions of the execution environment and security specification. The attestation may depend on the specific policy.

7. Attestation mechanism

Attestation happens in three situations:

 1) Grid user Alics needs to verify that the grid broker has corrected platform configuration before she submit her job;

 2) Grid broker finds an appropriate back-end node to run Alice's job. The attestation must be done to ensure that the target backend node is trusted;

 3) If one node is overloaded and needs migration, then the broker will exploit attestation to find another trusted backend node for migration.

8. Secure migration

Once the backend has no capacity to continually run the task, the job and it's execution environment will be migrated securely to another trusted backend node, which should be transparent to the end users.

Team Members

Hai Jin, Professor
Deqing Zou, Associate Professor
Feng Zhao, PH. D
Ge Cheng, PH. D. candidate
Weiqi Dai, graduate student
Jinjiu Long, graduate student
Lei Shi, graduate student
Yali Jiang, graduate student

Contact Information

Address
Service Computing Technology and System Key Lab & Cluster and Grid Computing Key Lab
School of Computer, Huazhong University of Science and Technology, Wuhan, 430074, China

Homepage
http://grid.hust.edu.cn/