Vehicular cloud research – What is missing?

Vehicular Clouds (VCs) have become an active research topic. However, even a cursory look reveals that the VC literature of recent years is full of papers discussing fanciful VC architectures and services that often seem too good to be true. And many of them are. It seems to us that promoting VC models without any regard to their practical feasibility is apt to discredit the VC concept altogether. Part of the problem stems from the fact that some authors do not seem to be concerned with the obvious fact that moving vehicles’ residency times in the VC may, indeed, be very short and, therefore, so is their contribution to the amount of useful work performed. Should a vehicle running a user job leave the VC prematurely, the amount of work performed by that vehicle may be lost, unless special precautions are taken. Such precautionary measures involve either some flavor of checkpointing or some form of redundant job assignment. Both approaches have consequences in terms of overhead and impact job completion time. The success of conventional cloud computing (CC) is attributable to the ability to provide quantifiable functional characteristics such as scalability, reliability and availability. By the same token, if the VCs are to see a widespread adoption, the same quantitative aspects have to be addressed here, too. Feasibility issues in terms of sufficient compute power, communication bandwidth, reliability, availability, and job duration time are all fundamental quantitative aspects of VCs that need to be studied and understood before one can claim with any degree of certainty that they can support the workload for which they are intended. The first contribution of this paper is to make a case for the stringent need to address quantitatively the performance characteristics of VC architectures and proposed services. Our second contribution is to point out directions and challenges facing the VC community.