Multi and Independent Block Approach in Public Cluster

MULTI AND INDEPENDENT BLOCK APPROACH IN PUBLIC CLUSTER Z. Akbar and L.T. Handoko Group for Theoretical and Computational Physic, Research Center for Physics, Indonesian Institute of Sciences, Kompleks Puspiptek Serpong, Tangerang 15310, Indonesia zaenal@teori.fisika.lipi.go.id ABSTRACT We present extended multi block approach in the LIPI Public Cluster. The multi block approach enables a cluster to be divided into several independent blocks which run jobs owned by different users simultaneously. Previously, we have maintained the blocks using single master node for all blocks due to efficiency and resource limitations. Following recent advancements and expansion of node's number, we have modified the multi block approach with multiple master nodes, each of them is responsible for a single block. We argue that this approach improves the overall performance significantly, for especially data intensive computational works. Keywords : cluster computer, resource allocation, multi block approach 1 INTRODUCTION LIPI Public Cluster (LPC) is globally a unique infrastructure due to its openness [1,2,3]. This nature leads to some innovations on cluster architecture, especially so called multi block approach to enable multiple blocks of small cluster running simultaneously without any interruption among each other [4]. In multi block approach, all running blocks have a common single master node as shown in Fig. 1. This is actually motivated by resource (especially hardware) limitations. For instance all nodes were not equipped with storage media. So, the initial runtime environment contains several daemons called MPD's in each block were booted disklessly through network using embedded boot ROM in network cards attached in each node. The master node further works as a gateway for users, and all blocks have only one MPD in it. Therefore this master node is the last point for users accessing the cluster. Some benefits in this approach are : It avoids possible overlapping or interruption among the nodes owned by different users. Number of nodes in an allocated block can be changed easily. It prevents anonymous accesses to another blocks owned by another users. Very efficient in initial construction and further maintenance works. As argued in previous paper [4], this technique is quite reliable and the overall performances are affected slightly. However, we concern that the result is valid as long as the node's number is small, namely at the order of few nodes. Moreover, the approach is suitable for some computational works that are processor (including memory) intensive, but not for the others which are data intensive. In some processor intensive jobs, the data traffic through network among the nodes during computation period is relatively small. Because once all sub-jobs predefined in a parallel programming were sent to and initiated at the allocated nodes, each of them is executed independently in a node almost without any communication with the others. In contrary, the case in some data intensive jobs like image mapping are quite different. This type of computational works usually require very intensive data exchange among the nodes during computation period. It is clear that then performance of cluster with conventional multi block approach would be decreased drastically in this case, since the master node is soon overloaded. In LPC currently the allocated nodes in a block is usually few, and the data intensive job is extremely rare. Because the facility is moreless used for educational and training field for beginners in parallel programming. However, we anticipate further advancements of our users and the increasing level of their jobs in the near future. Also, recently we have upgraded the hardware environment to be more sophisticated, i.e. all nodes are currently equipped with storage medias. So, regardless with some benefits in the conventional multi block approach, we are now involved in expanding the multi block approach with an independent master node in each block. In this paper, we first discuss the new approach, followed with the analysis performance before ending with conclusion. 2 MULTI BLOCK CLUSTER WITH INDEPENDENT MASTER NODES Now we are ready to discuss the extended multi block approach with an independent master node in each block. The most common architectures for cluster computer are the symmetric and asymmetric clusters [5]. In the symmetric cluster all nodes are treated equally and accessible by external users. In contrast, there is only one node that is accessible by external users and performs as mediator between users and the rest nodes in the asymmetric cluster. In this case, there should be a public interface for user and another private interfaces for nodes. In the LPC with independent master nodes, we

…(Full text truncated)…

This content is AI-processed based on ArXiv data.