optimization for scyld environment
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Vann H. Walke walkev at presearch.comWed Oct 30 05:26:34 PST 2002
- Previous message: optimization for scyld environment
- Next message: Cluster System for Computer Graphics Rendering
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Sashi, I'd really suggest reading through some of the available documentation. At http://www.beowulf.org you can find some good (if getting a bit dated) information. Robert Brown's web page for the Duke University "Brahma" cluster has quite a bit of information and links to other sources: http://www.phy.duke.edu/brahma/ . And there are a number of books available on building Beowulf / Linux clusters. Any of these sources will do a much better job discussing the design and function of a Beowulf than I will here. But, I'll do my best to provide some quick answers... On Wed, 2002-10-30 at 00:37, sashidhar wrote: > Hi, > I am new to linux cluster environments. Recently we have ported some C programs to a > linux cluster of 3 nodes running the Red Hat Scyld OS. Currently, our programs do > not employ any special parallel libraries like PVM or MPI. > > Does Scyld parallelize the programs by default, atleast upto some extent without > using the specialized libraries? Scyld does not parallelize by default. Scyld provides an integrated distribution that has all the libraries and kernel modifications to allow the writing and running of parallel programs easily. It does not parallelize for you. I don't know of any system, Scyld or otherwise, that will do an effective job of automatic parallelization. To use the parallel capabilities of a Beowulf cluster you have two main options: a) start a normal serial job on multiple nodes with different arguments / inputs. b) design your program to take advantage of the parallel environment (most commonly done with MPI or PVM). An alternative system for running lots of serial jobs is Mosix. Mosix does what is sometimes called "fork and forget." New processes are automatically migrated onto slave nodes to run. Note that this isn't parallelization of an individual program. It just provides an easier method of doing (a). > > Can I still use GCC -O / -O1 / -O2 .. flags to compile these files. Does this still > increase the performance or does this have any side effects? Sure. Optimization flags will optimize, just don't expect to get any parallelization out of them. The gcc used in Scyld is the same gcc used in other environments. Generally they flags "should" provide better performance without side effects, but if you have some strange code optimization can cause problems. But, this is more of a compiler issue which you should be able to find plenty of information on the web. Good luck, Vann Presearch, Inc.
- Previous message: optimization for scyld environment
- Next message: Cluster System for Computer Graphics Rendering
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
