(em português)

2K Resource Management Service

Introduction

The main purpose of this service is to provide sharing of resources among dynamic sets of distributed users. It should allow consistent and inexpensive access to the resources regardless of their geographic location.

This work is an extension of the resource management system developed in 2K. 2K is a distributed adaptable operating system. It adresses the problem of resource management in heterogeneous systems, providing support for automatic configuration, quality of service, dynamic reconfiguration, monitoring, etc.

Resource Management in 2K

The Resource Management Service in 2K is built as collection of LRMs(Local Resource Managers), GRMs(Global Resource Managers) and CRs(Component Repositories).

Usually, a set of hosts are grouped forming a cluster on which 2K is installed. Each host in the cluster has a LRM, that notifies the GRM periodically about significative changes in the availability of resources on that host, e.g., memory and CPU load. Moreover, it offers a distributed program execution service.

The GRM is an extended CORBA Trader that provides some additional features. Its responsibility is to maintain and consolidate information about resource availability provided by the LRMs. It also forwards component execution requests to others LRMs, choosing the remote host that has the best conditions to offer the quality of service requested by a component.

This service was implemented by Tomonori Yamane and can be downloaded with the 2K/dynamicTAO standard distribution.

Extension of the model

The previous version of the resource management service in 2K is appropriate when the objects are in the same cluster. They are isolated from resources available outside their domains. However, it might be that an object requires an amount of memory that is not available in any machine in the local cluster but that is available in some remote cluster.

Thus, a mechanism that expands their domains and aggregates more resources in a single wide-area, distributes system would be very useful. It would increase the probability of finding resources that fit the the needs of user applications.

Tipically, a cluster is composed of 2 to 100 hosts, each one with its LRM. Each cluster has one GRM. For clusters to cooperate sharing resources, they require an infrastructure to allow sharing of information about resource availability across multiple clusters. Thus, the main purpose of this work is to introduce and develop a mechanism to link different clusters.

Our concern in the design of this extension to the 2K Resource Management Service is to guarantee the scalability of the system. It is also necessary to add and remove clusters with a minimal impact over the system.

Clusters are linked using a tree structure to share information. Two or more domains that want to offer their resources are linked through other GRM data, which give hints about the state of the resources on each one of its direct descendents.

Our current implementation can be found at 2k-08.09.2004.tgz (instable version). More versions to come soon.

Experiments plan (in portuguese).

Related articles

• Gerenciamento de Recursos Distribuídos em Sistemas de Grande Escala (Distributed Resource Management in Large-Scale Systems). Jeferson Roberto Marques and Fabio Kon. 20th Brazilian Symposium on Computer Networks (SBRC'2002). (PPT Slides)
• The Design and Implementation of The 2K Resource Management Service. Tomonori Yamane, MS Thesis, Department of Computer Science, University of Illinois at Urbana-Champaign. February, 2000.

Point of contact: jmarques@ime.usp.br Do not hesitate to write me!