28 Aug RESEARCH NOTE: Are Wimpy Cores Good for Brawny Storage?

When could “wimpy” cores beat brawny ones? They have potential to do so in large-scale distributed storage deployments. This paper explores performance and resiliency tradeoffs enabled by using Calxeda’s ARM-based EnergyCore processors and their fabric-based system level architecture as the underpinning for a Ceph distributed object store implementation. Calxeda, Inktank, and Penguin announced a partnership in June, 2013 to deliver a software-defined storage appliance based on Calxeda’s EnergyCore processors and hardware architecture, Inktank’s Ceph distributed storage software architecture, all on Penguin’s UDX1 servers.  Based on performance comparisons run by Calxeda, Ceph scales well using Calxeda’s processors and system architecture. Calxeda is currently working with customers and partners to create a performance measurement framework for comparison to x86 storage servers.

Table Of Contents

  • Executive Summary
  • Distributed Storage Background
  • Distributed Storage Systems
  • Ceph Background
  • Calxeda EnergyCore and Ceph
  • Ceph Performance Using Calxeda EnergyCore Processors
  • Call to Action
  • Resources
  • System Configuration Used for Performance Analysis
  • Sidebar: Penguin on Demand POD
  • Figure 1: Total System Throughput by Node Count
  • Figure 2: Total System Throughput by Drives per Node
  • Figure 3: System Throughput per Node
  • Figure 4: System Throughput per Drive

You can download the paper here.