1 2 3 4 5 Previous Next 93 Replies Latest reply: Oct 27, 2009 12:09 PM by Billy~Verreynne Go to original post RSS
      • 30. Re: Finding appropriate block size?
        108476
        I think I'm right in saying that nobody knows.
        I asked someome at Oracle, and they strongly DENIED that they take ever take shortcuts when testing their software.

        And I don't believe it either.

        Testing is simple and straightforward, and Oracle hires some of the best software engineers on the planet, graduates of MIT, Yale, Harvard . . .

        Oracle does indeed have many bugs related to non-standard blocksizes, and yes, it makes Oracle look bad, especially to the competition.

        But, in fairness, Oracle has apologized, and is working out the bugs . . .
        • 31. Re: Finding appropriate block size?
          635471
          burleson wrote:
          Hi Jonathan,

          Rich Niemiec rejected all of your suggestions too. He didn’t "fix" a single one of the mistakes that you noted, not one . . .
          Would that be because Mr Niemich believes that the criticism is not valid, or because the book had already been printed and that there was no subsequent edition to make a correction in? Was there any public discussion of this in which JL's points were refuted?
          • 32. Re: Finding appropriate block size?
            635471
            burleson wrote:
            Please remove my book content from your post.

            It's copyrighted, not covered by fair use, and you do not have my permission to copy my work, ever.
            IANAL but I believe that "fair use" only applies to copyrighted works, and permission is not required. In fact if permission was given then fair use would be moot.
            • 33. Re: Finding appropriate block size?
              Billy~Verreynne
              Don Lewis wrote:

              So, can I take it that, on everage, straying from 8k is wrong?
              No. I said that not using 8KB block size has to be accompanied with justification and sound technical reasoning why 8KB block size is not ideal, and the chosen block size is.
              I ask as ,like many shops, we use 32k for our data warehouse. We have SAN and huge SGAs all on AIX 10g.
              And your technical reason for that is.. what?

              On some SANs, publishing the largest possible RAID10 LUNs comes with a 128KB block size on the SAN side. And when dealing with a TB database, you want a few large LUNs than 100 or so tiny 16GB LUNs.

              So what block size is then optimal read for the SAN? 128KB? And if used on the tablespace size, what will the impact be on the db instance and buffer cache? So in this case it is SAN performance vs. db impact.. and the balance between the two.

              Performance tuning is about understanding that balance.
              • 34. Re: Finding appropriate block size?
                635471
                burleson wrote:
                I think I'm right in saying that nobody knows.
                I asked someome at Oracle, and they strongly DENIED that they take ever take shortcuts when testing their software.

                And I don't believe it either.

                Testing is simple and straightforward, and Oracle hires some of the best software engineers on the planet, graduates of MIT, Yale, Harvard . . .

                Oracle does indeed have many bugs related to non-standard blocksizes, and yes, it makes Oracle look bad, especially to the competition.

                But, in fairness, Oracle has apologized, and is working out the bugs . . .
                Nobody is saying that Oracle took shortcuts, Don. However everyone acknowledges that any but the most trivial software is not 100% testable for all possible inputs and conditions, and it is not unreasonable for Oracle to focus testing on the default configuration. Based on that it is probably true that the default options are the most extensively tested, and therefore we are less likely to experience bugs if we stick to those defaults.

                Not universally true, especially when the default changes from a long-established mechanism to a relatively new one (eg. automatic memory management issues), but not unreasonable.
                • 35. Re: Finding appropriate block size?
                  108476
                  IAMAL
                  Oh, that's right, they call them solicitors in your country.

                  Data Warehouse Engineer, expert in intellectual property, you are truly legendary . . .
                  • 36. Re: Finding appropriate block size?
                    108476
                    Would that be because Mr Niemich believes that the criticism is not valid
                    You would have to ask Rich, but he is not one of Lewis' fans, any more than I am . . .
                    the book had already been printed and that there was no subsequent edition to make a correction in?
                    No, it's been redone for 11g, and it's an amazing book.

                    It's the bestselling Oracle tuining book in the world, and rightfully so, despite the "poor" rating by Lewis.

                    There is not a single proof nor test case in the whole book!
                    • 37. Re: Finding appropriate block size?
                      108476
                      However everyone acknowledges that any but the most trivial software is not 100% testable for all possible inputs and conditions
                      Everyone except Jonathan Lewis, Richard Foote and the "Oracle Scientists" . . .

                      They believe that a a SINGLE test case "proves" things . . .

                      Even more outlandish, they believe that a single negative test case "proves" that something is completely wrong!
                      and it is not unreasonable for Oracle to focus testing on the default configuration.
                      No, it's not reasonable at all, and I don't believe it.

                      Many of the Oracle defauilts are not right for everybody.

                      A "default" is just a "holding space", there is nothing that suggests that ANY default value is "best".

                      Take all_rows for example.

                      It may be the default, but most OLTP databases want a first_rows optimization.

                      Are you really suggesting that Oracle would test less on first_rows optimiaztion, just because it is not the default optimizer_mode?
                      • 38. Re: Finding appropriate block size?
                        635471
                        burleson wrote:
                        Data Warehouse Engineer, expert in intellectual property, you are truly legendary . . .
                        I've also been called "Oracle guru", "a well-respected data warehouse consultant".
                        • 39. Re: Finding appropriate block size?
                          635471
                          burleson wrote:
                          Would that be because Mr Niemich believes that the criticism is not valid
                          You would have to ask Rich, but he is not one of Lewis' fans, any more than I am . . .
                          It's rather difficult to draw conclusions from a lack of changes if you do not know why changes weren't made then, no?
                          No, it's been redone for 11g, and it's an amazing book.
                          I don't see it listed on amazon yet ... are you connected with its publication in some capacity?
                          • 40. Re: Finding appropriate block size?
                            108476
                            "Oracle guru", "a well-respected data warehouse consultant".
                            And rightfully so.

                            Some of your blog posts are nothing short of amazing . . .
                            • 41. Re: Finding appropriate block size?
                              635471
                              burleson wrote:
                              and it is not unreasonable for Oracle to focus testing on the default configuration.
                              No, it's not reasonable at all, and I don't believe it.
                              "Focus" does not mean "fail to test on anything else", it just means that if you are testing a new feature like 11g's new partitioning options then you might not choose to test it on 2kb block sizes as much as you test it on 8kb block sizes. I would not be surprised if tests were done, nor would I be surprised if fewer tests were done. A choice has to be made.

                              >
                              Many of the Oracle defauilts are not right for everybody.
                              Of course -- that's why not all parameters are hidden.
                              A "default" is just a "holding space", there is nothing that suggests that ANY default value is "best".
                              Not best for every case, but I think you need a rationale to depart from them.

                              >
                              Take all_rows for example.

                              It may be the default, but most OLTP databases want a first_rows optimization.

                              Are you really suggesting that Oracle would test less on first_rows optimiaztion, just because it is not the default optimizer_mode?
                              They probably do very little testing on first_rows because it has been deprecated for some time. However, I don't think that every test query is run under both all_rows and the first_rows_n optimizations, particularly ones that are intended for testing functionality rather than the optimizer. Are you suggesting that every query is run against objects on every possible blocks size and on every possible optimizer mode? How many test executions do you have to do there to cover just those two variables?
                              • 42. Re: Finding appropriate block size?
                                635471
                                burleson wrote:
                                "Oracle guru", "a well-respected data warehouse consultant".
                                And rightfully so.

                                Some of your blog posts are nothing short of amazing . . .
                                Thank you.
                                • 43. Re: Finding appropriate block size?
                                  Jonathan Lewis
                                  burleson wrote:

                                  Please remove my book content from your post.
                                  It's copyrighted, not covered by fair use, and you do not have my permission to copy my work, ever.
                                  Further to my earlier review of your book, I note that if you go down a few more pages from [+*the link supplied by Charles Hooper*+|http://books.google.com/books?id=bxHDtttb0ZAC&pg=PA28#v=onepage&q=&f=false] you will find (page 31) a section with the title "Materialized Views" that displays some figures that are supposed to be from a statspack report from 9i.

                                  This reviewer is not impressed with an example of a "Top 5 Timed events" that shows:
                                  <ul>
                                  db file scattered reads totalling 3,246 second and accounting for 82.04 percent of the time while ...
                                  db file sequential reads total 7,146 seconds, appear two lines further down the report, and account for 4.54 percent of the time.
                                  </ul>
                                  Apart from the inconsistency of the numbers, this reviewer also notes that the CPU time has an entry under the "Waits" column for "CPU time". In a genuine statspack report, this would be blank.

                                  This reviewer is most unimpressed that 17 months after the author of the book was made aware of these errors he still chose to reprint them.

                                  Regards
                                  Jonathan Lewis
                                  http://jonathanlewis.wordpress.com
                                  http://www.jlcomp.demon.co.uk

                                  To post code, statspack/AWR report, execution plans or trace files, start and end the section with the tag {noformat}
                                  {noformat} (lowercase, curly brackets, no spaces) so that the text appears in 
                                  fixed format
                                  .
                                  
                                  
                                  "If you are +*commenting upon*+ or +*critiquing*+ a copyrighted work--  +*for instance, writing a book review*+ -- fair use principles allow you to reproduce some of the work to achieve your purposes"
                                  http://fairuse.stanford.edu/Copyright_and_Fair_Use_Overview/                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                        
                                  • 44. Re: Finding appropriate block size?
                                    706417
                                    +"If you are commenting upon or critiquing a copyrighted work-- for instance, writing a book review -- fair use principles allow you to reproduce some of the work to achieve your purposes"+
                                    http://fairuse.stanford.edu/Copyright_and_Fair_Use_Overview/

                                    Stanford is not the last word in law, but statute and case law is: http://www.copyright.gov/fls/fl102.html

                                    Interestingly, something that is taken into consideration is:
                                    1. The purpose and character of the use, including whether such use is of commercial nature or is for nonprofit educational purposes.

                                    It gets a little hairier here: http://www.copyright.gov/title17/92chap5.html#506
                                    § 506. Criminal offenses
                                    (a) Criminal Infringement. —

                                    (1) In general. — Any person who willfully infringes a copyright shall be punished as provided under section 2319 of title 18, if the infringement was committed —

                                    (A) for purposes of commercial advantage or private financial gain;

                                    Your long-standing, well & truly documented and famously unpleasant ding-dong with Don would, without doubt, spell trouble if Don ever did get off his chair (where he sits "buck naked") and actually bothered to call his brief. I suspect you'd go deathly pale and have an accident if you did find yourself up before the beak. And, if it all went horribly wrong, you could (in this crazy age) end up never being admitted to America again. We know how crackpot they are about legal tangles at border control areas. Some poor folk have been held for hours, without so much as a decent cup of tea, and then sent back home (cargo class, I imagine) for traffic violations, etc.

                                    Sleep tight! And best of British luck to you!

                                    Regards - Don Lewis
                                    1 2 3 4 5 Previous Next