Virtual Application Appliances in Practice: Basic Mechanisms and Overheads

Virtual application appliances (VAA) (i.e., prebuilt virtual machines (VM) for specific scientific applications) are useful mechanisms to deal with the packaging of complex software systems and heterogeneous software environments (e.g., library version conflicts on different clusters and clouds). As an experience paper, we discuss some basic techniques for creating VAAs (e.g., virtual disk repositories (VDR)) and scripting their execution. As an evaluation paper, we quantify some of the key overheads, including the cost of staging data into/out of the VAA and the costs of VM migration. Subsequently, we introduce and evaluate a new Copy Over Shared Memory (CSM) mechanism to reduce the stage in/out overheads of data using secure, shared-memory regions between the host and guest machines. Our empirical evaluation shows that VAAs achieve nearnative, end-to-end performance in widely used bioinformatics applications that we tested (i.e., GROMACS, GAFolder, HMMer). We focus on data movement, VM boot up, shutdown and migration overheads of VAAs and find that they are negligible with respect to total run-times.