a computer cluster, usually simply cluster, of English. cluster = grape/cluster, bundle, swarm, mentioned, marks a number of interlaced computers, which in many cases as a computer to be seen from the outside to be able. Usually are the individual elements of a clusteramong themselves connected by a fast network. A goal of the “Clustering” mostly exists in the increase of the computing speed or the availability opposite an individual computer. The computer in a cluster (also knot or server) server farm are also often called.
Table of contents
the term cluster actually describes primarily architecture thatindividual components and their cooperating. One differentiates between so-called. homogeneous and heterogeneous Clustern. Computers homogeneous cluster run under the same operating system and same hardware, with the heterogeneous cluster can different operating systems or hardware be used. Well-known Linux cluster software are e.g.: “Beowulf” or “OpenMosix”.
Cluster become fora number of often different intended purposes assigned:
high availability cluster (hectar)
high availability cluster (English. High Availability - cluster) become the increase of the availability and/or. used for better reliability. Footstep on a knot of the cluster an error up, the services running on this cluster become on another knotmigrated. 2 knots possess most hectar cluster. Cluster exists with those on all knots of services constantly runs, this cluster active-actively calls one and/or. symmetrically or hot standby. If not all knots are active, one speaks of active passive or asymmetrically. Both the hardware and the softwarea hectar cluster must freely of single POINT OF failures (components, which would bring the entire system due to an error to the loss) its. Apply such hectar cluster in critical environments, in which maximum down-times are permitted by only few minutes in the year. In the context of disaster scenarios critical computer systems must securedbecome. In addition the cluster knots are often platziert apart several kilometers in different computing centres. In the emergency the knot can transfer the entire load in the computing centre not concerned. One calls this kind of Clustern also „stretched cluster “.
load Balancing cluster
load Balancing cluster become the purposethe distribution of load on several machines developed. The distribution of load is usually made by a redundantly laid out, central instance. Possible operational areas are environments with high requirements at computer performance. The performance need is covered here not by armament of individual computers, but by adding additional computers. Reason forthe use is not least the use of low-priced standard computers (COTS components) instead of of expensive special computers.
High performance Computing (HPC) cluster
HPC cluster serve for the processing of arithmetic problems. These arithmetic problems are divided on several knots. Either the tasks become in„Portions “divided and parallel on several knots implemented or the arithmetic problems (jobs called) are distributed on the individual knots. HPC cluster often are within the scientific range.
the first cluster product available in the trade was ARCnet, which 1977 of DATA POINT developbecame. The company DEK had first real success in the year 1983 with the conception of the product VAXCluster for its computer system VAX/VMS. The product did not only support parallel counting on the cluster knots, but also the common use of file systems and devices of all knots involved.These characteristics are not this very day contained with many free and commercial products.VAXCluster is available as VMS cluster also today still from the company HP for the operating system OpenVMS and the processors alpha and Itanium.
The failover function is usually made available by the operating system (Servicefailover, IP assumption). The assumption of services knows z. B. by the automatic migration by IP addresses or using a multicast address to be reached.
Generally differentiated between architectures shared emergency-hung and shared all.
Typical representative „active active " - clusters with shared emergency-hung architecture is DB2 with EEE (speak “tripolarize i”). Here each cluster knot accommodates its own data partition. A performance gain is obtained by the Partitionierung of the data and the distributed processing accompanying with it. Reliability is not ensured hereby.
Different is thiswith „shared all " - the cluster. This architecture ensures competitive access to Shared by a STORAGE that all cluster knots can access the entire volume of data. Apart from scaling and performance increase by this architecture also an additional reliability is reached. If a knot precipitates, the other knots take over its task (n).A typical representative shared of all architecture is material Oracle Application cluster (RAC).
Services must be particularly for a cluster programmed. A service is called cluster aware, if it on special events (like z. B. the loss of a cluster knot) and this reacts in more suitablyWay processes.
Cluster software can be implemented both in the form of Scripten in addition, be integrated in the Betriebsystemkernel.
with HPC Clustern is often divided the task which can be settled, the job, by means of a Decomposition program into smaller parts and then on thoseKnot distributes.
Communication between job parts running on different knots usually happens by means of Message Passing interface MPI, since fast communication between individual processes is desired. In addition one couples the knots with a fast network such as z. B. InfiniBand.
A usualMethod for the distribution of jobs on a HPC cluster is a job Scheduling program, which can make a distribution after different categories like z. B.Load sharing Facility (LSF) or network Queueing system (NQS).
- Heiko Bauke, Stephan Mertens: Cluster Computing. Springer 2006, ISBN 3-540-42299-4
- Charles Bookman: Linux Clustering. ISBN 1-578-70274-7
- Hartmann Gebauer: Clustering with Windows NT. Addison-Wesley 1999, ISBN 3-827-31403-8
- Andrea hero: Oracle 10g high availability. Addison-Wesley 2004, ISBN 3-827-32163-8
- Michael Soltau: Unix/Linux high availability. MITP 2002, ISBN 3-826-60775-9
- Martin Wieczorek, Uwe Naujoks, Bob Bartlett (Hrsg.): Business Continuity. Springer 2003, ISBN 3-540-44285-5
Web on the left of
- TOP500 list of the highest performance supercomputers
- cluster from 12 VIA EPIA800-Rechnern
- MPI - The Message Passing interface (MPI) standard
- wackamole - hectar cluster software (shifts IP addresses)
- heartbeat - hectar cluster software
- the Beowulf project - distribution counting
- open Mosix - distribution counting
- Kimberlite - Hectar-FailoverCluster
- HACMP - Hectar cluster software of IBM for AIX (English)
- MC/Service Guard - hectar cluster software of HP for HP-UX and Linux
- VMSCluster - VMSCluster of HP for OpenVMS
- Oracle RAC - cluster software of Oracle (English)
- SunCluster - cluster software of Sun Microsystems (English)
- Windows 2003 Clustering services hectar cluster software by Microsoft
- cluster server of VERITAS (Symantec) for AIX, HP-UX, Linux (talk has & Suse), Solaris and Microsoft (W2K & Windows 2003)