## Workstation Cluster Computation from the Perspective of the User (1994)

Venue: | Institut fur Informatik, Universitat Zurich |

Citations: | 3 - 0 self |

### BibTeX

@TECHREPORT{Cap94workstationcluster,

author = {Clemens H. Cap},

title = {Workstation Cluster Computation from the Perspective of the User},

institution = {Institut fur Informatik, Universitat Zurich},

year = {1994}

}

### OpenURL

### Abstract

. The use of workstation clusters for solving demanding application problems has widely been adopted. This paper analyses structural aspects which influence the performance of algorithms on such "hypercomputers". Two structurally very different examples help estimating the reasonable range of quantitative measures. An overview on organizational and infrastructural requirements answers the question who and where clusters presently may be used successfully. The paper concludes with recent trends and future developments. 1 Introduction The typical RISC workstation has 20--200 MIPS and 1--20 MFLOPS processor performance, 16--128 MB main memory, 0.5--5 GB disc capacity and 1--5 MB/s bandwidth for disc I/O. It runs a UNIX flavored operating system and costs from 5.000$ to 100.000$. Presently employed communication technology is the ethernet, with faster media just being introduced to the market. Today's RISC workstations therefore are able to cope with problems which eight years ago could ...

### Citations

1053 | Condor - a hunter of idle workstations - Litzkow, Livny, et al. - 1988 |

119 |
The string to string correction problem
- Wagner, Fisher
- 1974
(Show Context)
Citation Context ...o the role of the new macromolecule [18]. This similarity search is performed by a dynamic programming algorithm, which determines those strings in a data base which have a short string edit distance =-=[31]-=- to a given string. This algorithm can be parallelized by decomposing the data base of known molecules. Since there are more than 100.000 sequenced genes, the degree of possible parallelism is extreme... |

96 | User's Guide to the P4 Parallel Programming System
- Butler, Lusk
- 1992
(Show Context)
Citation Context ...s that 90% of Lawrence Livermore National Laboratory's workload could be solved with this technology. There is already a large number of research prototypes like PVM [15], Parform [10], Picl [14], P4 =-=[6]-=-, Parmacs [17], Lips [25] or POSYBL [27], and there are even commercialised solutions like LINDA [1] or Express [23]. Examples demonstrate, that for special applications the hypercomputer is a cheaper... |

84 |
Applications experience with Linda
- Carriero, Gelemter
- 1988
(Show Context)
Citation Context ...board activity or load measurements indicate the presence of such processes. Grace to its automatic load balancing mechanism, Linda and its derivates [1] are the preferred software platforms. Piranha =-=[5]-=- is a project exploiting expecially idle workstations, Parform [10] aims at automating load balancing for specific application problems. Such nondedicated clusters may be found at the universities of ... |

68 |
PICL: A portable instrumented communications library, C reference manual
- Geist, Heath, et al.
- 1990
(Show Context)
Citation Context ... estimates that 90% of Lawrence Livermore National Laboratory's workload could be solved with this technology. There is already a large number of research prototypes like PVM [15], Parform [10], Picl =-=[14]-=-, P4 [6], Parmacs [17], Lips [25] or POSYBL [27], and there are even commercialised solutions like LINDA [1] or Express [23]. Examples demonstrate, that for special applications the hypercomputer is a... |

62 | A Survey of Software Environments for Exploiting Networked Computing Resources
- Turcotte
- 1993
(Show Context)
Citation Context ...buted memory parallel computer [9], [10]. With this technique a performance or more than 1 Gflops [16] and 20 Gips [8] can be reached, exploiting the computing power of otherwise idling workstations. =-=[30]-=- estimates that 90% of Lawrence Livermore National Laboratory's workload could be solved with this technology. There is already a large number of research prototypes like PVM [15], Parform [10], Picl ... |

51 |
The PVM system: Supercomputing level concurrent computations on a heterogeneous network of workstations
- Geist, Sunderam
- 1991
(Show Context)
Citation Context ...idling workstations. [30] estimates that 90% of Lawrence Livermore National Laboratory's workload could be solved with this technology. There is already a large number of research prototypes like PVM =-=[15]-=-, Parform [10], Picl [14], P4 [6], Parmacs [17], Lips [25] or POSYBL [27], and there are even commercialised solutions like LINDA [1] or Express [23]. Examples demonstrate, that for special applicatio... |

48 |
Efficient parallel computing in distributed workstation environments
- Cap, Strumpen
- 1993
(Show Context)
Citation Context ...performance of the individual workstations to form a (virtual) parallel computer or hypercomputer , allowing to program a network of workstations just like a distributed memory parallel computer [9], =-=[10]-=-. With this technique a performance or more than 1 Gflops [16] and 20 Gips [8] can be reached, exploiting the computing power of otherwise idling workstations. [30] estimates that 90% of Lawrence Live... |

34 | The ANL/GMD macros (PARMACS) in fortran for portable parallel programming using the message passing programming model - users' guide and reference manual
- Hempel
- 1991
(Show Context)
Citation Context ... Lawrence Livermore National Laboratory's workload could be solved with this technology. There is already a large number of research prototypes like PVM [15], Parform [10], Picl [14], P4 [6], Parmacs =-=[17]-=-, Lips [25] or POSYBL [27], and there are even commercialised solutions like LINDA [1] or Express [23]. Examples demonstrate, that for special applications the hypercomputer is a cheaper [22] and fast... |

21 |
Express Version 1.0 : A Communication Environment for Parallel Computers
- Corporation
- 1988
(Show Context)
Citation Context ...ady a large number of research prototypes like PVM [15], Parform [10], Picl [14], P4 [6], Parmacs [17], Lips [25] or POSYBL [27], and there are even commercialised solutions like LINDA [1] or Express =-=[23]-=-. Examples demonstrate, that for special applications the hypercomputer is a cheaper [22] and faster [8] choice than a traditional supercomputer, and international prizes have been awarded for the per... |

18 | The Art of Distributed Applications - Corbin |

13 |
Linda on Distributed Memory Multiprocessors
- Bjornson
- 1991
(Show Context)
Citation Context ...l component seq m (p), which now depends on p, for estimating the behaviour of an application, when further increasing the number of processors: seq m (p) = (p \Delta T (p) \Gamma T (1))=(p \Gamma 1) =-=(4)-=- The sequential component may be given in absolute units as seconds run time and in relative units as percentage of the total run time T (p). The difference between actual parallel runtime and ideal p... |

13 |
On the parallel solution of 3D PDEs on a network of workstations and on vector computers
- Griebel, Huber, et al.
- 1993
(Show Context)
Citation Context ... parallel computer or hypercomputer , allowing to program a network of workstations just like a distributed memory parallel computer [9], [10]. With this technique a performance or more than 1 Gflops =-=[16]-=- and 20 Gips [8] can be reached, exploiting the computing power of otherwise idling workstations. [30] estimates that 90% of Lawrence Livermore National Laboratory's workload could be solved with this... |

12 | Parallel molecular sequence analysis on workstations in the Internet
- Strumpen
- 1993
(Show Context)
Citation Context ...e up the tree. No further communication is necessary. For our test runs we chose two different configurations. Measurements in a local area network consisting of a variety of machines are reported by =-=[29]-=-. A speedup curve scaled to Sparcstation1 units is reproduced in figure 5. 0 10 20 30 40 50 60 70 number of sparcstation1 units 0 10 20 30 40 50 60 70 s p e e d u p . . . . . . . . . . . . . . . . . .... |

10 |
A Proposal for a User-Level, Message-Passing Interface in a Distributed Memory Environment
- Dongarra, Hempel, et al.
- 1992
(Show Context)
Citation Context ... still remains an open problem. Although PVM plays the role of a de facto standard for the user and the software developer and even is evolving to an official message passing interface (MPI) standard =-=[13]-=-, there is a lack of official standards or even agreed concepts at higher levels of abstraction for programming. We already explained above, that hypercomputers presently are unstable systems. Further... |

9 |
DQS users guide
- Revor
- 1992
(Show Context)
Citation Context ...s a cost effective and reasonably powerful substitute for supercomputers. In such environments PVM [15] or similar systems are used as hypercomputer software, together with batching software like DQS =-=[24]-=- for scheduling purposes. Organizational and load balancing problems are easily solved in such environments. Clusters of this kind are used for example at the universities of Zurich, Vienna, Graz, Inn... |

9 |
LiPS: a System for Distributed Processing on Workstations
- Setz, Roth
- 1992
(Show Context)
Citation Context ...ivermore National Laboratory's workload could be solved with this technology. There is already a large number of research prototypes like PVM [15], Parform [10], Picl [14], P4 [6], Parmacs [17], Lips =-=[25]-=- or POSYBL [27], and there are even commercialised solutions like LINDA [1] or Express [23]. Examples demonstrate, that for special applications the hypercomputer is a cheaper [22] and faster [8] choi... |

6 |
Computational approaches to discovering semantics in molecular biology
- Lipton, Marr, et al.
- 1989
(Show Context)
Citation Context ...boratory, data bases are searched for similar genes and proteins with an already known biologic function. The results of these similarity searches may give a clue to the role of the new macromolecule =-=[18]-=-. This similarity search is performed by a dynamic programming algorithm, which determines those strings in a data base which have a short string edit distance [31] to a given string. This algorithm c... |

4 |
Network Supercomputing: A distributed-concurrent direct SCF scheme
- Luthi, Almlof
- 1993
(Show Context)
Citation Context ...ising clusters becomes more and more available, also the hypersupercomputer consisting of serveral interconnected vector- and supercomputers is studied for extremely compute intensive problems. [26], =-=[20]-=-. The National Science Foundation NSF presently is integrating their centers for studying cluster computing at the supercomputer level [21]. Workstation clusters also provide excellent solutions for s... |

3 |
ROBINHOOD: resource sharing by time stealing between DOS PCs on a LAN
- Asawa, Misra
- 1991
(Show Context)
Citation Context ...odel, the (theoretical) run times T t (p) with p processors and the sequential and parallel components satisfy equation (1) T t (1) = seq t + par t (1) and for all p we have T t (p) = seq t + par t p =-=(2)-=- We obtain equation (3), which holds for all values of p. Since seq t denotes a constant, the right side of (3) takes identical values, independent of the choice of p. seq t = p \Delta T t (p) \Gamma ... |

3 |
Issues on the Implementation of Programming Systems for Distributed Applications
- Shoinas
- 1991
(Show Context)
Citation Context ...al Laboratory's workload could be solved with this technology. There is already a large number of research prototypes like PVM [15], Parform [10], Picl [14], P4 [6], Parmacs [17], Lips [25] or POSYBL =-=[27]-=-, and there are even commercialised solutions like LINDA [1] or Express [23]. Examples demonstrate, that for special applications the hypercomputer is a cheaper [22] and faster [8] choice than a tradi... |

2 |
et al., SHIFT: The Scalable Heterogeneous Integrated Facility for HEP Computing
- Baud
(Show Context)
Citation Context ... run times T t (p) with p processors and the sequential and parallel components satisfy equation (1) T t (1) = seq t + par t (1) and for all p we have T t (p) = seq t + par t p (2) We obtain equation =-=(3)-=-, which holds for all values of p. Since seq t denotes a constant, the right side of (3) takes identical values, independent of the choice of p. seq t = p \Delta T t (p) \Gamma T t (1) p \Gamma 1 (3) ... |

2 |
Massively Parallel Computing in Wide Area Networks
- Cap
- 1993
(Show Context)
Citation Context ...r or hypercomputer , allowing to program a network of workstations just like a distributed memory parallel computer [9], [10]. With this technique a performance or more than 1 Gflops [16] and 20 Gips =-=[8]-=- can be reached, exploiting the computing power of otherwise idling workstations. [30] estimates that 90% of Lawrence Livermore National Laboratory's workload could be solved with this technology. The... |

2 |
Architekturelle Entscheidungen zur Unterstutzung parallelen Rechnens in Workstation Clustern und deren Realisierung
- Cap
(Show Context)
Citation Context ... the performance of the individual workstations to form a (virtual) parallel computer or hypercomputer , allowing to program a network of workstations just like a distributed memory parallel computer =-=[9]-=-, [10]. With this technique a performance or more than 1 Gflops [16] and 20 Gips [8] can be reached, exploiting the computing power of otherwise idling workstations. [30] estimates that 90% of Lawrenc... |

2 | The Parform -- A High Performance Parallel Platform for Efficient Computation in a Distributed Workstation Environment - Cap, Strumpen - 1992 |

2 |
Massively Parallel Computing in the Internet -- Entry to the SuParCup '93
- Cap, Strumpen
- 1993
(Show Context)
Citation Context ...22] and faster [8] choice than a traditional supercomputer, and international prizes have been awarded for the performance of workstation cluster solutions (Gordon Bell prize 1992 [22], SuParCup 1993 =-=[12]-=-). The hypercomputer is beginning to establish itself for certain classes of applications as an interesting alternative to the supercomputer. However, it will never be and never was intended to be an ... |

2 |
NSF pursues computing without walls
- Marsan
- 1992
(Show Context)
Citation Context ...s is studied for extremely compute intensive problems. [26], [20]. The National Science Foundation NSF presently is integrating their centers for studying cluster computing at the supercomputer level =-=[21]-=-. Workstation clusters also provide excellent solutions for special applications where mainly integer operations are used. The combined fast integer units of clustered RISC workstations can easily sur... |

2 | Netzwerk-Computing. Diploma thesis, Institut fur wissenschaftliches Rechnen - Sprenger - 1992 |