A good part of Clustrx® was created from scratch to break critical limitations of legacy monitoring and management systems (such Nagios, Ganglia, Torque, etc) and create really easy and comfortable environment for end users' access to supercomputers.
Strong accents were made on better utilizing of computing power and better manageability of supercomputing systems in full range of sizes, from micro-clusters to PFLOPS-class systems.
Clustrx® ships as integral installation package that contains:
- OS for master node (modified Linux + administration GUI)
- Set of drivers and patched Linux kernels to support specific hardware set
- Computing node's OS image generator and deployment manager
- Cluster management system, includes configuration database
- Cluster monitoring subsystem
- Task management system
- Accounting and billing subsystem
Cluster management system is the backbone of Clustrx® which serves distributed configuration database and distributed network services (DHCP, TFTP, etc) to support fast system deployment and jointless integration between all parts of the system. This system provides easy administration GUI for system deployment and support.
Cluster monitoring system was built from scratch to support polling, logging and processing the data from 1.5M data sources (hardware sensors and software data sources like CPU loads, kernel parameters, etc) per second on 1U physical monitoring server (which's enough to monitor 20k-25k of computing nodes). Monitoring servers can be scaled linearly by adding physical servers to monitoring farm. Such architecture is the base to build cluster's life support system, including emergency shutdown subsystem and modular business logic to specify cluster wide automated actions (such energy saving shutdown, etc). Logging capabilities give the base to build business logic of accounting subsystem.
Task management system is tightly integrated with monitoring subsystem to support dynamic rule-based tasks re-scheduling and advanced queue management in heterogenous computing environments (in case if cluster has GPU or Cell BE parts).
Accounting subsystem allows system administrators to implement any required policies to define users' rights, roles and tasks scheduling.
Clustrx® is scalable to manage hundreds PFLOPS of computing power per cluster without any changes in current system architecture, the set of kernels shipped with each installation can be adapted to support any required hardware including heterogenous computing environments. Legacy HPC application software created for Red Hat Enterprise Linux (including 32-bit packages) should run on Clustrx® without any modifications.
Clustrx® is registered trademark of Massive Solutions Ltd (Gibraltar) and shall be accessible as product through T-Massive Computing Limited in Oct 2009. T-Massive Computing is the future joint venture company of Massive Solutions Ltd (Gibraltar) and T-Platforms (Cyprus), it shall be fully established in 2009. Co-operation established between Massive Solutions and T-Platforms allowed to support all T-Blade II advanced features in Clustrx® distribution (all new service networks, utilizing chassis' control power, advanced power management, etc)