Performance analysis of HPC applications in the cloud

Abstract

The scalability of High Performance Computing (HPC) applications depends heavily on the efficient support of network communications in virtualized environments. However, Infrastructure as a Service (IaaS) providers are more focused on deploying systems with higher computational power interconnected via high-speed networks rather than improving the scalability of the communication middleware. This paper analyzes the main performance bottlenecks in HPC applications scalability on Amazon EC2 Cluster Compute platform: (1) evaluating the communication performance on shared memory and a virtualized 10 Gigabit Ethernet network; (2) assessing the scalability of representative HPC codes, the NAS Parallel Benchmarks, using an important number of cores, up to 512; (3) analyzing the new cluster instances (CC2), both in terms of single instance performance, scalability and costefficiency of its use; (4) suggesting techniques for reducing the impact of the virtualization overhead in the scalability of communication-intensive HPC codes, such as the direct access of the Virtual Machine to the network and reducing the number of processes per instance; and (5) proposing the combination of message-passing with multithreading as the most scalable and cost-effective option for running HPC applications on Amazon EC2 Cluster Compute platform.

MPI applications' performances in native vs. virtualized environments using InfiniBand IPoIB virtualization and live migration

“…CPU cycles, spent for TCP/IP processing cannot be used for application processing which led to an increasing drop in performance as we were adding more nodes. Similar results were obtained in related work [10] on 10 GbE virtualized interconnects. GROMACS application on the DPPC membrane system uses a less intensive MPI communication pattern.…”

“…Early benchmarks reports low reliability and performance, outlining network performance issues in [8] with further analysis in [9] showing improved but still low performance compared to native systems. Other studies evaluating HPC in the Cloud [10,11] highlight the benefits of on-demand and elastic resources and on the other hand the drawbacks of non-optimal performances, mainly due to interconnect limitations, brought by a virtualization layer and in some cases due to lower performance servers, used in general-use public cloud environments.…”

Optimising an Agent-Based Model to Explore the Behaviour of Simulated Burglars

“…Even after some simplifications from the original configuration (Malleson et al, 2010b), a single model run still required approximately 10 hours to complete on a normal desktop machine. The results discussed here were generated with the use of a 16-core Intel Xeon E5-2670 ("Sandy Bridge") virtual machine provided through Amazon Web Services (Expósito et al 2013), but even with this hardware, each GA iteration --with a population of only 20 chromosomes --required approximately 20 hours to complete. An associated side effect of the computation time is that it is not feasible to run each individual model configuration multiple times, which would be preferable because it would give a more comprehensive assessment of the model error (the simulation is probabilistic so each run will lead to slightly different results).…”

Resource Management for HPC on the Cloud

“…Yet, we list it here since it represents one of the few projects dealing with on-demand HPC. 11 offers three products: dedicated systems (R-Edge), on-demand HPC computing with technical support (R-Fusion), and pay-per-use customizable HPC resources (R-Cloud). R-cloud is an IaaS providing both a shared cluster, in which users are billed by the job, and virtual private clusters, in which access is billed on a 24/7 basis.…”