The Next Generation ClusterCyberinfrastructure Computing Management Standard SMASH

1
The Next Generation Cluster/Cyberinfrastructure
Computing Management Standard -- SMASH
Yung-Chin Fang, Tong Liu Dell Inc.
Chokchai Leangsuksun Louisiana Tech University
Stephen L. Scott Oak Ridge National Laboratory
2
TOC

• Motivation
• HPCC/HA Scale Out Fact
• The Problem
• The Facts
• The Idea
• The Action
• The Flexible Architecture
• Result

3
Motivation

• Driven by demand
• The successfulness of HPCC/HA cluster computing
  and grid computing research were driven by
  Federal budget
• FY06 Federal planned 2.2 B for supercomputing
  research
• Cyberinfrastructure computing is a focus, HPCC/HA
  scaling out is a fact
• When a node in Grid is down, user have to inform
  the administrator to reboot the box. This makes
  fully automation to be difficult.
• Hardware independent management framework is the
  needed foundation for Cyberinfrastructure
  Framework to achieve successfulness

4
HPCC/HA Scale Out Fact Manageability Need
5
The Problem
Dell Management Framework
Weak Interoperability Manageability
IBM Management Framework
Cray Management Framework
Intel Management Framework
SGI Management Framework
HP Management Framework
Ad Hoc Management Framework
De Facto Management Framework
6
The Facts

• A unified management architecture/framework is
  needed for productivity improvement
• All venders to drop their established framework
  to work on a new management architecture/framework
  is not likely to happen
• Research institute and academia to develop a
  management framework to cover all hardware
  architecture and management framework is also not
  likely to happen
• This hardware level manageability and
  interoperability can make Cyberinfrastructure to
  achieve higher degree of successfulness

7
The Idea

• Create an industry standard on top of well
  established architecture/framework so venders
  will buy in
• Flexibility for research institute and academia
  to facilitate the architecture, enhance/modify
  implementation and import/export data to fit
  their need
• Compatibility for current and future management
  architecture/framework
• No extra overhead (memory footprint/CPU cycle) on
  all compute nodes

8
The Action

• Distributed Management Task Force (DMTF) forms a
  working group to investigate this idea.
• The workgroup is called Server Management Working
  Group SMWG
• DMTF/SMWG initiates the Systems Management
  Architecture for Server Hardware (SMASH)
• SMASH development member includes Dell, IBM,
  Intel, HP, etc.
• SMASH is a suite of specifications that deliver
  architectural semantics, industry standard
  protocols and profiles to unify management
  environment

9
TheFlexible Architecture
10
TheFlexible Architecture
There are more than one way to implement the
architecture to achieve the same desired result
11
Result

• Enhance deployment and operational phase
  operation efficiency
• Provide hardware independent RASS manageability
• Reduce the total cost of ownership
• Set the foundation for heterogeneous
  Cyberinfrastructure (consists of HACC/HA
  clusters) management
• Enables info-structure-load sensitive resource
  manager
• Enables scientist to investigate even larger
  scale problems and bring higher precision result
  faster than ever.

12
Backup

• SMASH CLP -- System Management Command Line
  Protocol specification
• Example
• Promptgt show display associations o
  formatclpxml /system1/powersup3
• SMASH can input/output text/XML/CIM for binary
  portability and data exchangeability

13

• SMASH Profiles. Server Management profiles
  provide the object model definitions for
  manageability content and architecture models for
  mapping computer hardware to fully connected
  association graphs
• SMASH Managed Element Addressing Specification.
  This specification defines the formulation of
  command target addresses that resemble
  hierarchical file system naming conventions.
• SMASH CLP-to-CIM Mapping Specification. This is
  defined to facilitate existing CIM based
  management console
• SMASH CLP Discovery
• 1) How a client discovers which managed elements
  the MAP manages,
• 2) Discovering the capabilities of the MAP
  itself,
• 3) Discovering the service access points of the
  MAP. The MAP is a network-accessible service for
  managing a managed system

14
FIN
SMASH THIS