Español
English
Deutsch
      
       
  Engineering    ::   Supercomputing
 
 
Clusters
One of the most significant tendencies in calculation in the new millennium is the replacement of enormous monolithic servers by more flexible solutions based on clusters.

Clusters of low price servers, based on Linux-run PCs are now the reference in performance and scalability.

We are proud to present Englobe’s solutions in High Performance Computing (HPC), which as Myrinet did, are becoming the reference for clients who need high performance solutions.
  F-18

 


Physically, a PC cluster is a standard network, which instead of using a 100 Mb Ethernet topology, uses a network of much greater velocity (up to 2+2 Gbs) among each pair of PCs that compose it.

The processing capacity of the machines in cluster configuration and their minimal bandwidth requirement for interconnection permits creation of a server capable of processing large amounts of information and large quantities of calculations that were hitherto unattainable by individual machines. 

The novelty is that a cluster of PCs permits us to exploit the interconnection among the machines for specific applications that would make excessive use of the processors.  The PC’s will instead work in parallel in the high speed network permitting us to carry out applications that before were only reserved for large servers, such as parallel calculation, processing of video and digital images, biotechnology... 

The capacity of process of the machines in cluster and their interconnection using a wide one of so large band permits to create a great capable servant of processing information and to carry out such quantity of calculations that were unattainable for our machines and network. As many clusters are built using already existing machines, the software that had previously been installed can continue functioning without modifications. This ensures that the current team of developers may continue working in a known platform.
The savings are significant, since the old network of lower bandwidth can still be used for less taxing client applications and operations such as database processing and electronic mail. The cluster connects to the already existing network through standard interfaces, which permits the remainder of the networks to have access to the data and applications of the cluster.

Englobe designs clusters using Linux and the most advanced technology in high-speed communications cards (Myrinet), with which it then configures the resultant network. This permits a significant savings of resources and money through the use of already existing platforms, supplanting purchase of large servers whose maintenance is a lot more expensive.

Example of cluster architecture (Escuela Técnica Superior de Ingenieros Aeronáuticos, Universidad Politécnica de Madrid).
      

Calculation Nodes:

48 dual Pentium IV 3.06 GHz nodes, 2GB RAM DDR per node (96 GB total),
diskless

Server (Front-End):

Dual Pentium IV 3.06 GHz Node, 2GB RAM DDR, 300 GB HD

Network:

Gigabit Ethernet Switch (1Gb/s), Myrinet 2000 (2 Gb/s)

Storage System:

NAS server connected with SCSI to 3 systems RAID
(RAID 5, 4TB; RAID 5, 1TB; RAID 0, 3TB)

Operating system:

Linux, kernel 2.6



Englobe works with:





Home | About us | Clients | Contact | Site Map

Englobe Technologies. © 2007. Tfno. +34 917 48 92 22