Link to this page
COMPUTING CLUSTERS
As of the construction of computing clusters, accumulated experience and know-how allows our experts for optimal machines and systems profiling for the needs of the Customer and His/Her applications.
Gained experience and know-how allows our specialists to profile cluster systems for Clients and their applications. We build computing clusters based on own design COTS toolboxes (nodes) or blade and rack-mount servers (ie. Dell PowerEdge, HP ProLiant BladeSystem, IBM). Offered luster systems are optimized in terms of computing power and internal network throughtput and latency. Usually interconnecting network relies on GbE UTP/STP, GbE/10GbE-FO, Infiniband, QuarkNet technologies with active networking devices provided by vendors such as Mellanox, Juniper, Cisco/Linksys, HP or 3Com.
If it’s agreed and wished by the Client we may install and test correctness of functioning of:
- commercial Client’s software (Ansys/Fluent, Abaqus, Simula, Catia, Paraview, Tecplot etc.)
- software developing platforms
- compilers: Intel, PGI, GCC, F90
- parallel computing libraries: : MVAPICH (MPI-1.2) MVAPICH2 (MPI-2), LAM-MPI, OpenMPI, MPICH
Power fault protection and uninterrupted functioning of the cluster is assured by designed individually for each machine UPS systems managed by master (head) node (or other power steering system). Usually convenient way of cluster management is realized by means of additional interconnecting subnet, OOB ports and/or KVM console switches.
In case of highly-loaded multi-user clusters we may install and configure batch systems/job schedulers (DRMS) such as Torque/OpenPBS, LSF and other.
As an information storage/repository we use ready-to-use NAS devices such as Q-NAP, Synergy or storage matrices of our own design, however for most demanding clients, large-scale clusters and mission-critical applications we can provide solutions based on FC attached storage systems and Lustre-based parallel distributed file system. For Lustre metadata servers (MDS) and heavy IO load and latency dependent points for data storing and retrieving we use fastest IO controllers and fast server-targeted SAS (15K RPM) and SSD 3.0 disks.
Additionally in the computing cluster and HPC domain we are able to perform tests in a form of synthetic tests (ie. linpack) as well as with real CFD/MES applications. Having a database of performance of other machines gives us ability to assess computational performance of tested system as well as possibility to compare it to other previously tested systems.
Very high specialization of those kind of machines does not allow to give a pricelist – price is set upon Client audit individually. If you are interested in building, modernization or consulting concerning clusters, clustering technologies, management, monitoring and othe aspects of HPC calculations with use of distributed computing machines (clusters in particular) please don’t hesitate to contact us at HPC@sssystems.eu