Browse Prior Art Database

COMPUTE ELEMENT CLUSTER FOR HIGH DENSITY DATA CENTER FAULT TOLERANT COST-EFFECTIVE SERVERS

IP.com Disclosure Number: IPCOM000244463D
Publication Date: 2015-Dec-14
Document File: 6 page(s) / 158K

Publishing Venue

The IP.com Prior Art Database

Abstract

A compute element cluster for high density data center, fault tolerant, and cost effective servers is disclosed. A compute element cluster is a dense, inexpensive rackmount chassis including dozens or even hundreds of discrete "compute elements." Each compute element is a discrete collection of a processor (CPU), local memory (RAM), ephemeral storage (local flash storage used for storing an operating system and local data), external non-blocking network connectivity, and out-of-band management. To keep costs down, each compute element has no upgradeability with everything permanently soldered including memory and the processor. Each compute element is designed with density and cost per core in mind at the expense of pure speed and modularity. Each compute element is independent of all other compute element; if components of a single compute element fail for any reason, it is simply disabled, leaving the remaining compute elements unaffected.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 26% of the total text.

COMPUTE ELEMENT CLUSTER FOR HIGH DENSITY DATA CENTER FAULT TOLERANT COST-EFFECTIVE SERVERS

ABSTRACT

[0001]               A compute element cluster for high density data center, fault tolerant, and cost effective servers is disclosed.  A compute element cluster is a dense, inexpensive rackmount chassis including dozens or even hundreds of discrete “compute elements.”  Each compute element is a discrete collection of a processor (CPU), local memory (RAM), ephemeral storage (local flash storage used for storing an operating system and local data), external non-blocking network connectivity, and out-of-band management.  To keep costs down, each compute element has no upgradeability with everything permanently soldered including memory and the processor.  Each compute element is designed with density and cost per core in mind at the expense of pure speed and modularity.  Each compute element is independent of all other compute element; if components of a single compute element fail for any reason, it is simply disabled, leaving the remaining compute elements unaffected.

                 

BACKGROUND

[0002]               There are a number of existing solutions for high density blade servers that either focus on high-end workloads (such as servers from Dell and SuperMicro that provide 4 blades in a 2 Rack Unit (RU) form factor) or focus on low-cost modular workloads (such as HP's Moonshot project).  All existing solutions focus on modularity which increases cost.  In addition, there are no known solutions on the market today that reach the density required to allow practical separation of customer Virtual Machines (VMs) by hardware, which is necessary for many Network Function Virtualization (NFV) and Virtual Network Function (VNF) implementations.

DETAILED DESCRIPTION

[0003]               In various exemplary embodiments, the present disclosure relates to a compute element cluster for high density data center, fault tolerant, and cost effective servers.  A compute element cluster is a dense, inexpensive rackmount chassis including dozens or even hundreds of discrete “compute elements.”  Each compute element is a discrete collection of a processor (CPU), local memory (RAM), ephemeral storage (local flash storage used for storing an operating system and local data), external non-blocking network connectivity, and out-of-band management.  To keep costs down, each compute element has no upgradeability with everything permanently soldered including memory and the processor.  Each compute element is designed with density and cost per core in mind at the expense of pure speed and modularity.  Each compute element is independent of all other compute element; if components of a single compute element fail for any reason, it is simply disabled, leaving the remaining compute elements unaffected.

[0004]               The compute element cluster has individual compute elements on a same physical substrate (i.e., multiple discrete computers physically placed on the same system boa...