Introduction to the Special Issue

Abstract

One of the current trends in scientific cloud computing is that of migrating HPC to the clouds. This enables researchers to run their complex experiments on a pay-per-use basis without having to buy expensive datacenters. However it also raises some questions concerning the efficiency of cloud HPC and how cloud selection is made. Cloud computing is still in an early stage and solutions to these issues are currently being addressed. While some HPC applications such as MapReduce which does not require extensive intercommunication can be largely deployed on clouds, others like MPI jobs which require high speed networks are less suited for the average cloud infrastructure. To address this problem some cloud providers like Amazon have already begun offering virtualized clusters connected via high speed networks. The increase in the number of choices users must take makes selection a difficult subject especially due to the heterogeneous nature of the offers. To this aim an automatic platform for managing resources over multiple clouds for HPC is needed. These platforms must be able to broker user demands across several providers and meet specific objectives in terms of cost and time.

In this special issue we investigate these aspects and show some of the main contributions presented at dedicated workshops attached to the SYNASC 2013 conference. Namely we selected 4 papers from the Workshop on HPC Services,Workshop on Management of Resources and Services in Cloud and Sky Computing and Workshop on Agents for Complex Systems. A fifth paper presented during the main conference was also published in this number.

The first paper argues the necessity of HPC for computational intensive mathematical problems such as that of generating triangulations of PL 3- and 4-manifolds represented by edge-coloured graphs. The second paper addresses the problem of cloud interoperability and proposes a multi-PaaS application management solution. This provides a big step towards offering multi-cloud HPC services for problems such as the one previously mentioned. The third paper deals with benchmarking the efficiency of multi-cloud platforms. This step is crucial especially when using automated cloud platforms as it allows them to optimize their selection strategy based on real information on the underlying systems. The fourth paper presents an automated agent based negotiation model. While the agents are selfish they are also motivated by the necessity to cooperate with others for achieving their objectives. The model is well applicable to a cloud computing scenario where user agents need to negotiate for resources handled by provider agents. The final paper, which is not part of the special issue but has been presented at the main SYNASC conference track, deals with evolutionary structural testing. Emphasis is put on the study of the "switch'' statement in order to automatically generate test data for triggering a particular branch of the program. A new evolutionary structural approach based on a Compact and Minimized Control Flow Graph relying on two different formulas for evaluating the test data performance is proposed.