jueves, 3 de mayo de 2012

Supercomputers


A little introduction to supercomputers, with some of history

A supercomputer is the type of computer most powerful and fastest available at this time. How are you machines are designed to process huge amounts of information in a short time and are dedicated to a specific task, its application or use of the individual escapes rather engage in:
1. Search oilfield seismic large databases.
2. The study and prediction of tornadoes.
3. The study and prediction of anywhere in the world.
4. The development of models and projects the creation of aircraft, flight simulators.

We must also add that the supercomputers are a relatively new technology, so their use is not overcrowded and is sensitive to changes. It is for this reason that the price is very high with over 30 million dollars and the number of production per year is small. Supercomputers are the sort of more powerful computers and faster than existing in a given time. Are large, the largest among its peers. Can process huge amounts of information in a short time and can perform millions of instructions per second, are aimed at a specific task and have a very large storage capacity. They are also the most expensive having a costoque can exceed $ 30 million. Because of its high cost are produced very little for a year, there are even some that are made only on request.

They have a special temperature control in order to dissipate the heat that some components can reach. It acts as the arbiter of all applications and controls access to all files, so does the input and output operations. The user is directed to the organization's central computer processing support when required.
They are designed for multiprocessing systems, the CPU is the processing center and can support thousands of users online. The number of processors that can have a supercomputer depends mainly on the model, can have from about 16 to 512 processors (such as the NEC SX-4 1997) and more.


The Manchester Mark I
The first supercomputer British laid the foundation for many concepts still used today.
The digital world's first computer, electronic and program-loaded (top figure) successfully executed its first program on June 21, 1948. This program was written by Tom Kilburn who, along with the late FC (Freddie) Williams designed and built the Manchester Mark I computer This machine, the prototype Mark I, quickly became known as' The Baby Machine 'or just `The Baby'.
In modern terms The Baby had a RAM (random access memory) for only 32 positions or 'words'. Each word consisting of 32 bits (binary digits), which means that the machine had a total of 1024 bits of memory.
The RAM technology was based on cathode ray tube (CRT). CRTs were used to store data bits as charged areas on the phosphor screen, showing as an incandescent series of dots on it. The CRT's electron beam can efficiently handle this load and write a 1 or 0 and then read it as requested. Freddie Williams led the investigation that perfected the use of CRT storage with Tom Kilburn made decisive contributions.

The Cray 1
The Cray 1 was the first supercomputer "modern".
The first Cray-1 in England was located at Daresbury Laboratory for two years before being moved to the University of London.
The first major success in the design of a modern supercomputer was the Cray-1 which was introduced in 1976. One reason why the Cray-1 was so successful was that I could make more than one hundred million arithmetic operations per second (100 MFLOP / s). It was designed by Seymour Cray who left Control Data Corporation in 1970 to create his own company called Cray Research Inc., founded in 1972. If you today, following a conventional process, you will try to find a computer in the same speed using PCs, you would need to connect 200 of them, or you could just buy 33 Sun4s. Cray Reasearch Inc. made at least 16 of its fabulous Cray 1. A typical Cray-1 in 1976 cost over 700,000 dollars. The good thing is that you could instruct the machine in any color deseases, which still remains.

IBM - SP2 computer
The IBM SP2 installed on the Daresbury Laboratory, has, since an upgrade this year, 24 nodes P2SC (Super Chip), plus another 2 processors width oldest node located in two racks. Only the second of which is shown in the picture. Each node has a clock of 120 MHz and 128 Mbytes of memory. Two New High Performance Switch (TV3) are used to connect nodes together. Data storage is 40 GB of fast disk locally attached Ethernet and FDDI networks for user access.
An individual maximum node performance of 480 Mflops offers a total of over 12 Gflops for the entire machine.
A workstation is connected to PowerPC RS/6000 SP2 system for monitoring and management of hardware and software.

Bibliography:

http://www.monografias.com/trabajos65/supercomputadoras/supercomputadoras1.shtml
http://roberto.4mg.com/historia.htm
http://conecti.ca/2011/08/15/infografia-historia-de-las-supercomputadoras/

Semana 13

La aportacion de la semana fue una investigacion sobre el  algoritmo de mineria de datos, peticion de mi compañero avendaño, cabe destacar que se puede aplicar en clustering para hacer redes neuronales y donde me gustaria darle mas importancia a este algoritmo y donde mas se interreso mi compañero avendaño.

 link:
 http://elisa.dyndns-web.com/progra/Algoritmos/Mineriadedatos

mis nominaciones son para:
jonathan alvarado, Alejandor Avendaño y Osvaldo Hinojosa.

lunes, 30 de abril de 2012

Semana 12

La aportacion de esta semana fue hablar un poco sobre hamachi, que es y como funciona y un poco de su seguridad, cabe destacar que seria una muy buen herramienta para el cluster.

liga:
http://elisa.dyndns-web.com/progra/HamachiApplication

Mis nominaciones son para:
Alejandro Avendaño, Eduardo Triana y Adriana Contreras por la aprotacion:
http://elisa.dyndns-web.com/progra/Hamachi%2BSSH

Communication in distributed systems

The most important difference between a distributed system and a single processor systemis the communication between processes. In a single-processor communication implicitly assumes the existence of shared memory:
• Ex: problem of producers and consumers, where a process writes to a shared buffer and anotherprocess reads from it.
In a distributed system there is no shared memoryand thus the whole nature of communicationbetween processes must be rethought. Processes to communicate, they must adhere to rules known as protocols. For distributed systems over a wide area, these protocols often take the form of several layers and each layer has its own goals and rules. Messages are exchanged in various ways, thereare many design options in this regard, an important option is the "remote procedure call." It is also important to consider the possibilities of communication between groups of processes, not only between two processes. In distributed systems, the absence of physical connection between the different memories of the teams, communication is performed by message transfer.

The standard ISO OSI
To send messages using the standard ISO OSI (Open Systems Interconnection), a layered model for communication of open systems. The layers provide multiple interfaces with different levels of detail, the last being the most general. The OSI standard defines the following seven layers: physical, data link, network, transport, session, presentation and application. The OSI model distinguishes two types of protocols, connection oriented and connectionless protocols. In the first, before any data transmission requires a virtual connection, which must end after sending. Connectionless protocols do not require this step, and messages are sent as datagrams.

Asynchronous transmission mode ATMAsynchronous transfer mode or ATM provides a quick means of transmission. Higher speeds are achieved regardless of the information flow control and error control in the intermediate nodes of the transmission. ATM uses connection-oriented mode, and allows transmission of different types of information such as voice, video, data, etc..
The client-server model of communication based on a simplification of the OSI model. The seven layers that provides produce a waste of the transfer speed of the network, which will be used only three layers: physical, data link and request / reply. Transfers are based on the protocol request / response and eliminates the need for connection.
RPC
 
Group communication
Communication systems that support groups can be divided into two categories: closed groups andopen groups. In the first only allowed communication between group members,while in the latter processes that do not belong to the group can interact with it. Closed groups are often used, in general, parallel processing, where each member has its own objective and does not interact with the `` outside world''. Instead, open groups are intended for application developmentclient / server, where the processes interact with the group to apply for services.

We can also categorize groups of processestaking into account the role played by the membersthat compose them: peer groups and hierarchical groups. If in a group there is nodistinction between the processes that formed anddecisions are taken collectively, is said to be apeer group. However, if relations are establishedhierarchy in a group, where some processes have a greater say than others, we speak of hierarchical groups. Each of these organizations group has its advantages and disadvantages. In hierarchical groups, if they fail the processes that make decisions, can produce a mismatch of the other members for as long as this situation remains. Inthe peer group does not produce this effect, if it fails one of its members can continue working (witha smaller group).


Bibliography:
http://www.iuma.ulpgc.es/users/lhdez/inves/pfcs/memoria-ivan/node2.html#SECTION00240000000000000000

jueves, 19 de abril de 2012

semana 11

La aportacion de esta sema fue un poco de informacion de lo que se puede utilizar en pyMPI. Una complementacion de lo que hizo mi compañero Alejandro Avendaño.


Link:
http://elisa.dyndns-web.com/progra/pyMPI


Mis nominaciones son para:
Osvaldo Hinojosa
Alejandro Avendaño
Jonathan Alvarado

por las aportaciones de esta semana.

lunes, 16 de abril de 2012

Semana 10

El aporte de esta semana son unos programas en c del mergesort




y aqui el merge sort con threads no estoy seguro si esta hecho de una buena forma


Mis nominaciones son para:
Jose Gonzales, Alejandro Avendaño y Eduardo Triana

jueves, 5 de abril de 2012

semana 9

La aportacion de esta semana, hice referencia a los posibles fallos que puede presentar un sistema distribuido ya que como en esta semana se ve lo que es la tolerancia a fallos y en el laboratorio describi de manera general este tema.


El link es el siguiente:
http://elisa.dyndns-web.com/progra/Cluster


Mis nominaciones son a:
Jose Gonzales 
Jonathan Alvarado