The good folks at Basement Supercomputing have published a white paper comparing one of the company’s on-prem Limulus personal HPC appliance to Amazon EC2 Cloud instances. The economics are quite interesting.
In this careful study equivalent EC2 cluster instances were configured using current pricing (Spring 2019). A comparison of Basement Supercomputing Limulus appliance workstations with the Amazon EC2 (Elastic Compute Cloud) is presented. The capabilities of both approaches are discussed along with a detailed comparison of two Limulus appliance designs. The first appliance is a Limulus High Performance Computing cluster designed to work and operate in an Edge (non-data center) environment such as a classroom, lab, office, or factory floor. The second Limulus design is for a Hadoop/Spark Edge appliance. Each design was replicated as close as possible using Amazon EC2 instances.
A detailed comparison of both technical capabilities and price is presented resulting in the following conclusions:
- The Limulus appliance meets or exceeds the capabilities of a similar cluster configured using EC2 instances.
- The one-year lease cost for a comparably configured EC2 cluster equals or exceeds the purchase price of a Limulus cluster
- If the cost of renting EC2 instances for one year is applied to purchasing a Limulus appliance, the resource will provide at least 2-3 years of service.
- Software configuration and support costs are not part of the cloud instance cost. Limulus appliances are turn-key and require minimal administration effort.
appliance.
The Limulus workstation appliance is a true computation cluster consisting of 4-8 independent motherboards connected by Ethernet. All systems are designed to operate in an Edge (non-data center) environment such as classrooms, labs, offices, or factory floors. Using an optimized design, all Limulus systems provide consistent bare-metal performance coupled with a complete turn-key software suite for either High Performance Computing (HPC) or Hadoop/Spark applications.
Limulus systems are available with the following array of features:
- 4 to 8 motherboards with either Intel Core or Xeon processors (AMD in development)
- Up to 48 physical fast cores with 48 additional hyper-threads
- Up to 512 GBytes of Memory (ECC option available)
- Up to 64 TBytes of distributed SSD storage (used for Hadoop File System)
- Up to 56 TBytes of bulk centralized HDD storage
- Provides full internal 1 or 10 GbE Ethernet networking (The 10 GbE option has a reproducible TCP latency of 11 µseconds.)
- Uses a single residential/office/classroom wall outlet
- Operates quietly with little heat in an office/lab/classroom environments
- Is a turn-key ready to run appliance fully installed Linux-based HPC or Hadoop/Spark environments
- Has additional options, which include redundant power supply and a rack mount case
- Has full hardware and software support options
The full paper is available from Cluster Monkey (registration required). If you don’t want to register, you can Contact Basement Supercomputing directly and they gladly will send you a copy.
Sign up for our insideHPC Newsletter