Bright Cluster Manager
Advanced cluster management made easy Roland Richardson ClusterVision UK
[Link]
2010
Frankfurt
[Link]
2010
Cluster Installation
ClusterVision Europe Cl t Vi i provision and support clusters across E i i d t l t
UK, Germany, France, The Netherlands, Italy, Spain, etc. etc. Middle East US
Install OS, Configure Network drivers, install libraries and MPI implementations install execution daemons for WLM Some tasks are annoyingly different queue configuration, user admin hardware configuration Switches PDUs configuration admin, configuration, Switches, etc ClusterVision wanted to automate all those standard tasks while accommodating the variation and new features
[Link]
2010
The Toolkit approach pp
Most cluster management solutions use the toolkit approach (Linux distro + tools)
Examples: Rocks, PCM, OSCAR, UniCluster, xCAT, etc. Tools typically used: Ganglia Cacti Nagios Cfengine System Imager Ganglia, Cacti, Nagios, Cfengine, Imager, etc. Tools rarely designed to work together Tools rarely designed for HPC Tools T l rarely d i l designed t scale d to l Each tool has its own command line interface and GUI Each tool has its own daemon and database Roadmap dependent on developers of the tools
Issues with the toolkit approach:
Making a collection of unrelated tools work together
Requires a lot of expertise and scripting Rarely leads to a really easy to use and scalable solution easy-to-use
4
2010
[Link]
Annoying Differences y g
Each cluster deployed is unique
Different Applications Different Workload Different Infrastructure
For some this creates a complex management environment
No common practice Difficult to document
For so e it's fine o some s e
[Link]
2010
[Link]
2010
[Link]
2010
[Link]
2010
Architecture
CMDaemon
procedure call SOAP+SSL event
[Link]
2010
Bright CM Elements g
Cluster Management Daemon Cluster Management Shell Cluster Management GUI , y Linux distribution, with many enhancements
Red Hat & SUSE Enterprise CentOS & Scientific Linux
Node provisioning system Workload management system
Sun Grid Engine, Torque/Maui MOAB, PBS Pro, LSF
HPC user environment Cluster management & monitoring functionality Years of HPC expertise g Consistent integration of all the above ------------------------------------------------------------------------------ + Bright Cluster Manager
[Link]
10
2010
HPC User Environment
Let users focus on performing computations Rich collection of HPC software
Compilers (GNU, Intel*, Portland*, Open64, etc.) Parallel middleware (MPI libraries, threading libraries, OpenMP, Global Arrays, etc.) Gl b l A t ) Mathematical libraries (ACML, MKL*, LAPACK, BLAS, GOTO, etc.) etc ) Development tools (debuggers, profilers, etc.) Environment modules
NVIDIA CUDA, OpenCL & GPGPU drivers , p Intel Cluster Ready Compliant Compliant applications run out of the box
[Link]
11
2010
Bright Buzz: g
"It's hands-down the best GPGPU management software out there." - Joe Lipman, VP of Sales at Advanced HPC "With Bright Cluster Manager now offering full support for ScaleMP vSMP Foundation, setting up and managing a powerful SMP cluster has never been so easy " -- Shai easy. Fultheim, CEO, ScaleMP "Bright Cluster Manager allows us to provision thousands of nodes within hours and rapidly deploy software updates." -- D b d " Deborah N h Newman, Di Director Professional P f i l Services at PCPC Direct
[Link]
12
2010
Management Interface
Graphical User Interface (GUI) Offers administrator full cluster control Standalone desktop application Manages multiple clusters simultaneously R Runs on Li Linux, Wi d Windows, M OS X MacOS Built on top of Mozilla XUL engine Command Line Interface (CLI) ( ) All GUI functionality also available through Command Line Interface (CLI) Interactive and scriptable in batch mode
[Link]
Admin GUI
13
2010
[Link]
14
2010
[Link]
15
2010
[Link]
16
2010
[Link]
17
2010
[Link]
18
2010
[Link]
19
2010
[Link]
20
2010
[Link]
21
2010
Advanced Features
Advanced Features Redundant head nodes Cluster health checking g
Health checks before jobs are run Frequently scheduled health checks Hardware burn-in environment
Daemon with low resource consumption Multiple, load-balanced provisioning nodes
Allows scaling to thousands of nodes
N d di Node discovery using Eth i Ethernet switch port d t ti t it h t detection Live & incremental image updates A t Automated BIOS updates and configurations t d d t d fi ti
[Link]
22
2010
Unique Features q
Fundamental, integrated solution Complete management solution Scalable to thousands of nodes Minimal CPU overhead due to light weight daemon light-weight Support for redundant head nodes g , p g Image based, incremental provisioning Support for diskless and InfiniBand-only slave nodes Single GUI and CLI for all cluster management tasks Support f Red Hat Enterprise, SUSE Enterprise, CentOS and S for S S C OS Scientific Linux Cluster health checking Role based authentication Zoom-in function for monitoring graphs Visualization of metrics in Rackview Graphical parallel shell Fully graphical installation in 30 minutes
23
2010
[Link]
Questions? i
[Link]
24
2010
The End h d