clustering

cheap server cluster

Servers Cluster and storage cluster server

We offers a wide range of servers and storage to cluster. 
Basically, it's called cluster to a set of machines that work like a single system. 

There are different types of cheap server cluster: 
- Cluster supercomputing, HPC. With Gigabit networking, Infiniband, Myrinet. 
- Render Farm (Render Farm). It's a cluster specified to minimize the time rendering 
- Cluster Service. To ensure the altadisponibilidad 

List servers to cluster. Ask for a custom configuration. 

Server Cluster Supercomputing: 

In the environment of Supercomputing, all the machines (called nodes), working as one, making the calculus for all. It is a model widely used in centers of high-performance scientific computing and universities. It is a flexible type of cluster, you can mix types of nodes C2D, Dual Xeon, AMD Opteron, Itanium 2, uniprocessor, multiprocessor ... 
In this type of clusters, there are different distributions and modules, such as: 
Beowulf (PVM, MPI), openmosix ... 
Typically, communication between nodes is by Gbit Ethernet, but there are times you need more benefits and uses networking low-latency, the type of Myrinet, SCI, or the new networks based on PCI-Express Infiniband. Highlight of the Infiniband networks, which have the best latencies, and a bandwidth of up to 120GB / s.

These transfers are theoretical maximum, the real is 20% lower due to 8b/10b encoding. 

We deliver the complete turnkey solution, with all components necessary, wardrobe rack ensures that special refrigeration, KVM analog and / or digital connectivity, screen, keyboard and mouse in 1U, all structured in such a way that the administration system and add or replace components are made in a quick, easy and comfortable. 

Technical aspects of the Hardware: 

Wardrobe: 
The cabinet Rack is a crucial dimension of a cluster. Has a direct influence on issues such as refrigeration, accessibility, and the optimization of space. 
The refrigeration process directly affects the calculation, and an excess of temperature causes loss of performance, in fact at present most of the processors down the speed of the microprocessor when they detect excessive temperatures. Besides an adequate refrigeration prolongs the life of components beyond the warranty, which lowers the cost of maintaining the Cluster. 

Hosts: 
Wanted the best performance / price ratio, taking care issues such as the density and electric consumption. These past 2 above aspects are essential if we host the cluster in a data processing center, and we are going to charge for electric consumption and footprint. 
For the choice of server, you have servers with Intel and AMD microprocessors. In all cases we look for the platform that gives more benefits to the customer's application, and according to the agreements that we will ever even make a demo to see benefits and be quite sure that the platform chosen is the most appropriate. 
Currently the most sought after options are: 
- Servdores 1U Dual Xeon 
- Servers 1U Dual Core Xeon Quadruple 
- Servers 1U Quad Opteron Dual Core 
- Servers 1U Quad Opteron Dual Core 

Storage: 
These systems usually have a shared storage systems. In addition to the traditional RAID storage systems such as systems SAN or NAS, IP SAN (iSCSI), storage solutions have parallel object-oriented oriented supercomputing clusters object-oriented manufacturer's campaigns. 
This storage is characterized by multiple controllers working in parallel to achieve up to 600MB / s sustained transfer rate in per 10GbE ports. 

It should be noted that the networks with low latency and a cluster of stores can achieve improvements in performance up to 37 times working on applications such as Fluent. 

Low-latency networks: 

As alternative to the traditional Gigabit Ethernet, there is the possibility of using low-latency networks, which dramatically increase the benefits from a number of nodes. At present communication Gigabit Ethernet is a bottleneck for the calculation in parallel when we increase the number of nodes, which is why there comes a moment that leaves most cost-effective to implement a system of communication with low latency, which increase the number of nodes . 

At present we offer 2 different networks altencia low. Infiniband and Myrinet. The first gets the best of the day, while the second provides greater compatibility. 

Myrinet Cluster with: 
In some cluster, to optimize performance, we used high-performance networks for the communication between the nodes. In this case, Myrinet, is a system of communication, which have a microprocessor embedded in each HBA in charge of communications, which frees up labor to the cpu server. These networks are particularly appropriate when the cost per node is high, and the cost of a network Myrinet is a small percentage of the cost of a node (eg Cluster with Xeon, Opteron, Itanium 2). 

For such networks, using modular chassis, which are inserted cards of 8 ports. The last slot chassis, is reserved for a card monitoring. 

Management: 
We have different elements that facilitate the management of this type of installation. While it can be managed through SSH, greatly facilitates the work tecldo the inclusion of a folding screen and mouse, which occupies only 1U of rack inside. 
Additionally we offer KVM systems to monitor all nodes sharing a monitor keyboard and mouse, and even with access through IP with multiple users. 

List servers to cluster. Ask for a custom configuration. 

Render Farms

At present, due to the complexity and detail of 3D programs, it is necessary large amount of hours on a machine in order to render our work. On many occasions it is necessary to leave the machines running all night waiting for the results, thus making the delivery of the project is delayed dramatically. 
In addition to having a range of professional Workstation, we can mount a farmer or render "render farm" "Hosting, so that the process of rendering is not endless. 
These farms rendering normally consist of a parent or node management, which is a high-performance workstation with which it works, and connected by Gigabit Ethernet network, a server cluster rendering, which migrate toward the processes, which only mission is to be rendered perform. We have Dual Xeon server Woodcrest and Opteron Dual 1U in which we use indifferently in the light of that over-performance platform for implementing specific customer, with solutions up to 4 porcesadores Intel Xeon or AMD Opteron Dual-Core in two 1U servers. Do not forget the other components that need this type of system, besides getting maximum benefits, the minimum possible cost without neglecting issues such as consumption and space. 

Hosts: 
We seek the solution that's interesnate according to your needs. Wanted mainly related computing power / price ratio, making sure things like future expansion, budget, etc.. 

Wardrobe Rack: Depending on the number of nodes, we look for a cabinet that is to mold what you need. All our cabinets are professional range of processes for data centers, which ensures adequate cooling, and support up 1.500Kg weight. 

KVM: These systems are essential if not, facilitate the administration of servers. They enable the administration of all nodods from a single keyboard set-mouse-monitor. We have range Analog and digital spectrum, with remote access over IP with encryption of 128Bits, and up to 4 concurrent users. 

Try & Buy. 
To ensure the investment of your money, you provide the opportunity for AMD and Intel platforms to test to verify the performance, signing a contract for the Try & Buy. 

List servers to render. Ask for a custom configuration