HOURS: HarmOnic and Universal Resource Sharing
Foundations for Resource Sharing in Computing Communities  

Motivation

Shared Cyberinfrastructure (CI)—federated sharing of dispersed pools of geographically distributed computing resources under coordinated control—has been considered as a promising platform for solving large-scale problems in science and engineering, e.g., NSF's TeraGrid and DoE's Open Science Grid, for building the next generation of distributed Enterprise applications, e.g., open grid services architecture (OGSA), and for sharing resources in our daily lives, e.g., KaZaA, BitTorrent. This trend is a natural combination of “technology push” and “application pull,” and could lead to the emergence of a new service-oriented computing industry if it can be successfully deployed. On one hand, the low cost of COTS (Commodity-Off-The-Shelf) hardware makes it affordable to build a high-end cluster with reasonable performance. On the other hand, the ever-increasing multidisciplinary trend of most research fields demands collaboration among multiple computing sites.

We envision that there are five challenges related to resource sharing in such an open environment, including

•  Heterogeneity. The heterogeneity of resources makes the multiple-resource sharing difficult, because of lacking of a formal metric for the trading among different resources;
•  Untrustedness. Enforcing a cooperative, adaptive, and anti-maliciousness P2P sharing environment on top of an untrusted and private P2P community is really a challenge; 
• Selfishness. The possible threats launched by selfish peers, such as cheating and boasting, can destroy the cooperative resource sharing. Enforcing a fair resource sharing framework to limit the negative effect of the selfish peers is one of the goals for our approach;
• Autonomy and Cooperation. Peers usually belong to different administrative domains which may have different local policies. How to effective and efficient integration of these local policies and general resource sharing is a challenge;
  
• Incentives. Free riders are the considerable population in the P2P community. To attract the peer to contribute to the community is an old but still ongoing problem. In this project, we intend to address all these problems.
• Adaptiveness  The inherent dynamics of computing communities, e.g., joining/leaving the community or add/remove computing powers, necessitates the support of adaptiveness across all layers. 

Top

Our Approach

Choosing the good partner and avoiding the bad partner are the first but very important step for the resource sharing. Making use of the knowledge of the social network to build the trust inference model is our approach to achieve this goal. With the help of the trust model, every peer can build a neighbor set with high-quality peers, thus improve the reliability and stability of resource sharing. However, the trust model can not introduce more incentives, cooperativeness, and efficiency for the resource sharing. We borrow the idea from the real economics to build a currency-based model to complement the missing part of the trust model. Currencies play a role to bridge every units. People can use the currency to buy needed merchandises, or save it for the future use.  So the currency circulation is lite-version of the merchandise exchange. This phenomenon match the P2P system very well. The overview of our approach is showed in the following figure.

In the HOURS project, we propose an approach that consists of two models: M-CUBE, a Multiple CUrrency Based Economic model, as the decentralized trading scheme, and aPET, an adaptive PErsonalized Trust model, to provide the trustworthiness of the peer to support M-CUBE. The M-CUBE model provides a general and flexible substrate to support most of high level resource management services required by the P2P computing, such as resource coallocation, quality of service (QoS) control, advance reservation and scheduling algorithms. aPET is built on top of our previous work on PET, which derives the trustworthiness from the reputation evaluation and risk evaluation. The trustworthiness value provided by PET will be treated as the view of the peer by M-CUBE. The unique feature of our approach is seamless integrating the trustworthiness and dependability of peers into the resource trading.

Top

Projects

adaptive PET

Building a good cooperation in the P2P resource sharing is a fundamental and challenging research topic due to peer anonymity, peer independence, high dynamics of peer behaviors and network connections, and the absence of a perfect security mechanism of P2P systems. We propose PET, a personalized trust model to help the construction of a good cooperation, especially in the context of economic-based solutions for the P2P resource sharing. The trust model consists of two parts: reputation evaluation and risk evaluation. Reputation is the accumulative assessment for the long-term behavior, while the risk evaluation is the opinion of the short-term behavior. Two kinds of knowledge, interaction-derived experience (local knowledge) , and the recommendation (knowledge of other peers), are used to derive the reputation.  Selecting the weights of these two parts are environment specific, and is a decision on the trade off between the reliability and efficiency. In the P2P system, we are suggesting to put more weight on the reputation considering  high dynamics is a special characteristic in this kind of system.  The risk part is employed to deal with the dramatic spoiling of peers, which makes PET differ from other trust models based on the reputation only. Risk evaluation can make the trust model more sensitive. How to make sensitive match the correctness is our ongoing research.  This work contributes to first modeling the risk as the opinion of short-term trustworthiness and combining with traditional reputation evaluation to derive the trustworthiness in this field.  The figure-I is the description of the original PET model. Recently, we are working on an adaptive PET model that captures the dynamics of the system.

Figure-I Derivation of Trustworthiness Value

 

M-CUBE

M-CUBE, a multiple-currency based economic model, is a self-policing and distributed approach that is based on top of PET model, to lay a foundation for heterogeneous resource sharing in an untrusted P2P computing environment. With the help of the trust management and the merits of the economic institution, M-CUBE provides a novel self-policing and quality-aware framework for the sharing of heterogeneous resources, and is a flexible universal infrastructure for building high-level resource management related services.  M-CUBE is built upon currency-based mechanism, where the uniqueness of M-CUBE is each peer has its own currency. There are four major modules in M-CUBE: the Price Regulator decides the price of the resources; the Ratio Regulator determines the exchange ratio of the currency based on the trustworthiness value provided by the PET model; the Service Discovery module is in charge of discovering the available resources provided by remote peers; finally the Currency Exchange module enables peers to bargain until the agreement of the currency exchange is reached, and then makes the exchange.

• Ratings Analysis:  Ratings (also known as recommendations) from our family members, friends, and the people with high reputation are treated as reliable information sources, which are helpful and time-saving for human being to explore the quality of other people or services. Hinted by the human social network, the researchers are of great interest to employ ratings in the computing society. However, whether the ratings in the computing world has the same positive effect is still an open problem. In this subproject, we intend to answer the following questions:  Are ratings always helpful? How to configure factors related to the rating so as to improve the performance of the system? How the rating affected by other factors, such as the quality of raters, the scale of the system, the dissemination mode, and so on? and Are the complicated aggregation algorithms better than the simple ones?  We found some very interesting results, part of them is published in CollaborateCom 2005. The simulator of  different ratings algorithms comparison is available at Ratings-Simulator.
  

• SWAP: Data Management for Data-Intensive Scientific Applications The amount of scientific data generated by simulations or collected from large scale experiments have reached levels (e.g., peta bytes) that can not be stored in researcher's local computer center. In several applications, such as High Energy Physics, the data, which is stored in the back-end tapes, is fetched using an on-demand fashion by 100s-1000s of scientists all over the world. Such data are vital to large scientific collaborations dispersed over wide-area networks via either dedicated or commercial networks.  The access of data out of tape for analysis is usually the main bottleneck, rather than computing resources. In this project, we are investigating an efficient distributed data management scheme that intelligent swap the data between archived tapes and storage disks contributed by the computer centers. The goal is to maximize the resource utilization and throughput and minimize the makespan of scientific applications.  

• Workflow scheduling  Scheduling is the key to the performance of grid workflow applications where a collection of dependent jobs run on a heterogeneous and dynamic environment. Previous efforts in this area propose either static scheduling strategy which maps jobs to resources before execution time in order to achieve optimal performance of entire workflow, or dynamic alternative which schedules individual job only when it is ready to execute without considering workflow as a whole. While sizable work supports the claim that static scheduling performs better for workflow applications than the dynamic one, there have been arguments about whether this advantage can be really realized given the fact that grid environment changes constantly and the majority of real world grid workflow systems are implemented with dynamic strategy only. In this project, we investigate how to assure and exploit these intrinsic benefits of static scheduling strategies in grid environment and proposes a novel adaptive rescheduling concept, which allows the workflow planner works collaboratively with the run time executor and reschedule in a proactive way had the grid environment changes significantly. By adaptive rescheduling, a planner can better adapt to the grid dynamics, i.e., resource availability and performance fluctuation etc. An HEFT-based adaptive rescheduling algorithm is presented, evaluated and compared with traditional static and  dynamic strategies respectively.  We are currently  investigating the multiple-DAG scheduling in a failure-prone environment. 

• Peer-to-Peer Web Server Sharing  Peer-to-Peer Web server sharing is a case study of the HOURS project. The objective of this subproject is improving the reliability and availability of cooperative Web servers. The basic idea is a group of peers pool in their resources to help each other during individual peer's peak loads and/or system failures. The main concept behind the workability of this arrangement is an understanding that not all companies who form the peer to peer network will have peak loads on their web sites at the same instant of time. In fact if the companies are geographically dispersed spanning many time zones, the effectiveness of this system is more apparent.

Top

People

Jacqueline D. Brown

Zhengqiang (Sean) Liang 

Tung Nguyen

Dr. Weisong Shi   

Brandon Szeliga (former member)

Zhifeng Yu (former member)

Jayashree Ravi (former member)

Publications and Technical Reports

Top