Scientific Workflow Applications on Amazon EC2 (1005.2718v1)

Abstract: The proliferation of commercial cloud computing providers has generated significant interest in the scientific computing community. Much recent research has attempted to determine the benefits and drawbacks of cloud computing for scientific applications. Although clouds have many attractive features, such as virtualization, on-demand provisioning, and "pay as you go" usage-based pricing, it is not clear whether they are able to deliver the performance required for scientific applications at a reasonable price. In this paper we examine the performance and cost of clouds from the perspective of scientific workflow applications. We use three characteristic workflows to compare the performance of a commercial cloud with that of a typical HPC system, and we analyze the various costs associated with running those workflows in the cloud. We find that the performance of clouds is not unreasonable given the hardware resources provided, and that performance comparable to HPC systems can be achieved given similar resources. We also find that the cost of running workflows on a commercial cloud can be reduced by storing data in the cloud rather than transferring it from outside.

• The paper demonstrates that Amazon EC2 can effectively execute scientific workflows with a modest 1-10% virtualization overhead compared to traditional HPC clusters.
• The paper reveals that EC2’s on-demand pricing model offers cost efficiency for sporadic computations, despite trade-offs in I/O performance for data-intensive applications.
• The paper highlights that selecting appropriate EC2 instance types is crucial to match HPC performance for varied workflows in fields like astronomy, seismology, and bioinformatics.

Scientific Workflow Applications on Amazon EC2

The paper "Scientific Workflow Applications on Amazon EC2" thoroughly examines the potential and limitations of employing Amazon's Elastic Compute Cloud (EC2) for executing scientific workflow applications. Given the broadening interest in harnessing cloud infrastructure for scientific computations, this research explores the performance, cost, and feasibility of utilizing commercial cloud resources for scientific workflows, comparing them with traditional high-performance computing (HPC) systems.

Study Context and Methodology

The authors address the uncertainty about whether cloud services can meet the rigorous performance demands of scientific applications affordably, considering both technological and economic benefits. Through this lens, they assess three diverse workflows covering astronomy (Montage), seismology (Broadband), and bioinformatics (Epigenomics), chosen for their varied I/O, memory, and CPU demands. The experimentation contrasts EC2 resources against NCSA's Abe cluster, a typical HPC system, intending to determine performance parity under equivalent resource conditions.

For experimentation, the authors utilized several EC2 instance types alongside different storage configurations to employ single, multi-core nodes for performance appraisals. The selection of EC2 resource types embodies both computational and memory variants, accommodating tasks according to their distinct resource requisites.

Key Findings

1. Performance Analysis:
   • Differences in performance were noted across the workflow types contingent on their respective resource needs.
   • EC2 demonstrated reasonable performance with virtualization incurring between 1% to 10% overhead, a find consistent with preexisting virtualization research.
   • I/O-bound workflows like Montage showed superior performance on Abe due to its parallel file systems and high-speed interconnects, which EC2 lacks.
2. Cost Implications:
   • Resource costs predominated over storage and transfer expenses. EC2's on-demand pricing model aids economic efficiency, especially for sporadic computation needs.
   • For iterative or frequently accessed datasets, storing data in the cloud proved more cost-efficient than repeatedly transferring them, which aligns with storage cost dynamics versus transfer expenses.
3. Resource Utilization:
   • Memory-intensive applications require careful instance selection on EC2 to prevent performance bottlenecks due to inadequate memory allocations per core.
   • Broadband workflow illustrated near parity between EC2's c1.xlarge and Abe node resource usage when sufficient memory was provisioned.

Implications and Future Directions

This investigation imparts significant insights into the applicability of commercial clouds for scientific workflows. For many loosely-coupled scientific workflows, leveraging EC2's virtual infrastructure presents a viable alternative with notable flexibility and cost benefits, albeit with certain trade-offs concerning interconnect efficiency and I/O throughput capabilities.

From a practical perspective, these findings encourage the contemplation of cloud solutions for scientific computing within contexts not critically dependent on ultra-high-speed networks. The paper also discusses future potential for cloud providers to enhance competitiveness with traditional HPC systems, particularly through sophisticated network and storage configurations catering to I/O-intensive operations.

Looking ahead, the research suggests a continuation into multi-node scalability analysis, emphasizing communication efficiencies across cloud-deployed nodes. As cloud technology evolves, the landscape of scientific computing may witness shifts favoring flexible, scalable compute services offered by entities like Amazon EC2, given improvements in network infrastructure and parallel file systems.

The paper effectively demystifies various economic and technical aspects of cloud utilization, fostering a deeper understanding pertinent to current advancements and future explorations within the scientific computing domain.