1 2 Previous Next 26 Replies Latest reply on May 13, 2010 3:59 PM by timfox Go to original post
      • 15. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
        timfox

        We need some way of providing different defaults depending on whether AIO or NIO is used.

         

        Just having them in the config file is not sufficient, since as Clebert points out, if AIO is not available it will fall back to NIO with the wrong defaults.

        • 16. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
          clebert.suconic

          It's not just wrong defaults. Wrong values.

           

           

          The user must provide the values for AIO (if AIO is used) and NIO (if NIO is used).

           

           

          Using values meant for AIO on NIO could have collateral effects.

          • 17. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
            leosbitto

            Clebert Suconic wrote:

             

            It's not just wrong defaults. Wrong values.

             

             

            The user must provide the values for AIO (if AIO is used) and NIO (if NIO is used).

             

             

            Using values meant for AIO on NIO could have collateral effects.

             

            I do not like the possibility of silent fallback from AIO to NIO - that itself could have various collateral effects as well. Could it be possible for the administrator to disable such fallback? I.e. if AIO would not be available, complain loudly and do not start at all, instead of risiking NIO...

            • 18. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
              clebert.suconic

              Right now it will always fall to NIO sillently.

               

               

              You can request a feature (by opening a jira) such to enforce AIO and we could just throw an exception and stop startup if AIO is not available.

               

              But we can't change the default behaviour... as there will be users who think right the opposite.

              • 19. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
                timfox

                We also need the fallback so HornetQ/JBoss AS works transparently without reconfiguration on non Linux platforms where AIO is not available. This is a requirement from JBoss AS.

                • 20. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
                  leosbitto

                  Clebert Suconic wrote:

                   

                  Right now it will always fall to NIO sillently.

                   

                   

                  You can request a feature (by opening a jira) such to enforce AIO and we could just throw an exception and stop startup if AIO is not available.

                   

                  But we can't change the default behaviour... as there will be users who think right the opposite.

                   

                  Sure, here you are: https://jira.jboss.org/jira/browse/HORNETQ-360

                  • 21. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
                    leosbitto

                    Tim Fox wrote:

                     

                    We also need the fallback so HornetQ/JBoss AS works transparently without reconfiguration on non Linux platforms where AIO is not available. This is a requirement from JBoss AS.

                     

                    So what about configuring only NIO by default? That gives a consistent default behaviour. Anybody interested in the extra performance could configure AIO, of course. I get a bit scared every time I see software trying to be too smart, and such drastic automatic reconfiguration (AIO->NIO) sure did trigger that feeling for me.

                    • 22. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
                      timfox

                      Leos Bitto wrote:

                       

                      Tim Fox wrote:

                       

                      We also need the fallback so HornetQ/JBoss AS works transparently without reconfiguration on non Linux platforms where AIO is not available. This is a requirement from JBoss AS.

                       

                      So what about configuring only NIO by default? That gives a consistent default behaviour. Anybody interested in the extra performance could configure AIO, of course. I get a bit scared every time I see software trying to be too smart, and such drastic automatic reconfiguration (AIO->NIO) sure did trigger that feeling for me.

                      The issue there is performance.

                       

                      AIO can bet up to twice as performant as NIO on Linux platforms, and having great performance out of the box is a must, since, like it or not, many (most?) comparative benchmarks are done on default out of the box settings.

                      • 23. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
                        leosbitto

                        Tim Fox wrote:

                         

                         

                        The issue there is performance.

                         

                        AIO can bet up to twice as performant as NIO on Linux platforms, and having great performance out of the box is a must, since, like it or not, many (most?) comparative benchmarks are done on default out of the box settings.

                         

                        So it seems that the issue is not actually the performance, but the results of many (most?) comparative benchmarks, right? You cannot be sure what such ignorant benchmarks actually measure anyway - there is no guarantee that there would be properly working libaio available. Why not concentrate on those relevant benchmarks which actually do provide reasonable results? Those would have to read the documentation, look for some chapter about performance tuning, ...

                        • 24. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
                          clebert.suconic

                          I guess it's not possible to please everybody :-)

                           

                          We have chosen to provide automatic tuning. i.e. If AIO is available, use AIO, if not.. use NIO.

                           

                          Most users will not change those values anyway (timeout... etc).

                           

                          We will provide a AIO strict mode like you requested on the JIRA. (It's fairly simple to implement actually and change the docs).

                           

                          (Actually.. if someone is interested in contributing on this, feel welcome to jump in ....  :-) )

                          • 25. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
                            saguirre

                            I am new to this community, so please forgive me if I break any protocols in this post.  I see that https://jira.jboss.org/jira/browse/HORNETQ-296 is fixed in 2.1.0.BETA3.  Is there a scheduled timeframe for this to be included in a standard release?  Thank you.

                            • 26. Re: minLargeMessageSize effectively Ignored in 2.0.0 GA ?
                              timfox

                              Soon

                              1 2 Previous Next