Talk:LU reduction
From Wikipedia, the free encyclopedia
 |
|
| Stub | This article is a stub and has been rated Stub-Class on the assessment scale. |
[edit] Thanks
Hi Graeme thank you for accepting this article. I hope it is helpful for people who embark on performance monitoring and stumble upon this term exclusivley (I found) used by folks in supercomputing. It took me a while to figure out what it is. Though I am not sure LU reduction is simply a parallized version of LU decomposition (I haven't seen the algo) it is very likely so. After you read this, you can delete it ;-)
- I will leave this here as someone more knowlegeable can improve the definition. Graeme Bartlett 12:26, 20 September 2007 (UTC)
- I tried to find out what LU reduction actually means and amended the article. I'm also not 100% sure of it, but it is not simply a parallellized version of LU decomposition. The references I added has some source code. -- Jitse Niesen (talk) 12:55, 15 January 2008 (UTC)
- I just checked another paper which I added to the last reference and the code in the paper you added, and they both confirm my assumption that LU Reduction is a special LU decomposition algorithm in that it is parallelizable. Juitart's code (LUAppl) calculates L and U at the same time, distributing the task to different processors.Plumbum2 (talk) 02:17, 16 January 2008 (UTC)
- Sorry, I didn't realize you were still around, otherwise I would have talked to you first. I looked at Juitart et al.'s code again. I'm not sure about it (I didn't study the thread stuff in sufficiently detail), but it looks like you're right and that the code does compute both L and U. My impression after looking through the three papers is that LU reduction is used as a synonym for LU decomposition. LU decomposition is a problem (not a specific algorithm) and there are many algorithms for solving it, some of which are parallel; at least that's the usage in numerical analysis. However, supercomputers is not my field so if you think that LU reduction refers to a specific algorithm, then you're probably right. Anyway, I reverted my edit as I was wrong; thanks for checking. -- Jitse Niesen (talk) 16:45, 16 January 2008 (UTC)
- The paper you contributed helped me clear up the confusion. Well I don't think LU reduction refers to a specific algorithm. I actually compiled and ran the code with example matrices where I new the LU decomposition and looked at what the code produces. Hence, my conclusion. Also, only Juitart et. al refer to this term (what googled tells me anyway), so I basically wrote this article for others who may encounter the same puzzle while reading these guys' papers. They talk about LU reduction as if eating bread but never tell you what it means. 20:16, 21 January 2008 (UTC) Plumbum2 (talk) 14:07, 23 January 2008 (UTC)
- Sorry, I didn't realize you were still around, otherwise I would have talked to you first. I looked at Juitart et al.'s code again. I'm not sure about it (I didn't study the thread stuff in sufficiently detail), but it looks like you're right and that the code does compute both L and U. My impression after looking through the three papers is that LU reduction is used as a synonym for LU decomposition. LU decomposition is a problem (not a specific algorithm) and there are many algorithms for solving it, some of which are parallel; at least that's the usage in numerical analysis. However, supercomputers is not my field so if you think that LU reduction refers to a specific algorithm, then you're probably right. Anyway, I reverted my edit as I was wrong; thanks for checking. -- Jitse Niesen (talk) 16:45, 16 January 2008 (UTC)
- I just checked another paper which I added to the last reference and the code in the paper you added, and they both confirm my assumption that LU Reduction is a special LU decomposition algorithm in that it is parallelizable. Juitart's code (LUAppl) calculates L and U at the same time, distributing the task to different processors.Plumbum2 (talk) 02:17, 16 January 2008 (UTC)
- I tried to find out what LU reduction actually means and amended the article. I'm also not 100% sure of it, but it is not simply a parallellized version of LU decomposition. The references I added has some source code. -- Jitse Niesen (talk) 12:55, 15 January 2008 (UTC)
[edit] TODO
Cite template should be used for the references not as I did it now with references in brackets.Plumbum2 (talk) 02:37, 16 January 2008 (UTC)
