1 2 3 4 5 6 Previous Next 84 Replies Latest reply on Mar 30, 2015 6:23 AM by Cristian Vasile Go to original post
      • 45. Re: Would a RAM disk improve Server performance?
        Cristian Vasile

        Hello Matthew,

         

        25% less time on refreshing large extracts sounds good for me. Will be great if you could test SuperCache solution.

         

        Regards,

        Cristian.

        • 46. Re: Would a RAM disk improve Server performance?
          Toby Erkson

          Matthew, agreed, RAM disk doesn't appear to make much -- if any -- difference with smaller/faster extracts.

           

          Here's the script I'm using to 'turn on' the usage of the RAM disk:

          tabadmin set service.temp.dir F:\temp0
          tabadmin set temp.dir F:\temp1
          tabadmin configure
          

          I manually restart the server:

           

          Why the different "temp" directories?  Because I want to see what each does.  Currently running an extract and it uses the "temp0" folder.  "temp1" hasn't been created yet by the system.

           

          Confirmation: I cannot download workbooks when the RAM drive is in effect.  What gives?  As you can see, I gave 'Everyone' full access and I still get the 404 error:

          • 47. Re: Would a RAM disk improve Server performance?
            Toby Erkson

            Just an update on downloading a workbook RAM disk issue

             

            When an workbook download is requested it appears to go into the TableauTemp folder and the data goes into the Extracts folder (I watched that folder get populated):

            After the 404 error the Extracts folder disappears:

            • 48. Re: Would a RAM disk improve Server performance?
              Cristian Vasile

              Toby,

               

              you wrote ""temp1" hasn't been created yet by the system."

               

              F:\temp1 do exist? if not create it manually.

               

              Regards,

              Cristian

               

              P.S.

               

              ms dos tree /f  (tree /f  > tree.txt) command will display the entire directory and files structure in a beautiful layout, it's easy to copy / paste text versus printscreening.

              • 49. Re: Would a RAM disk improve Server performance?
                Toby Erkson

                F:\temp1 does not exist.  The system should create it, just like it did for F:\temp0.

                I did manually create that folder as you suggested and then tried to download a workbook and it still failed

                 

                I also got a 'disk full' error on one of my tests so I'm re-running that extract.  The extract was successful yesterday using the default Tableau temp values.

                 

                < my edit to your edit > I use Greenshot for screen capture/editing.  It's faster for me and provides exact visual layout for faster recognition by others.

                • 50. Re: Would a RAM disk improve Server performance?
                  Toby Erkson

                  Here's something interesting, my F: drive (the RAM disk) doesn't show up in my Windows Resource Monitor:

                  Nor does it show up in the Windows Disk Defragmenter:

                   

                  Maybe this is why the download workbook fails?  Just speculation.

                  • 51. Re: Would a RAM disk improve Server performance?
                    Cristian Vasile

                    Toby,

                     

                    Could you please publish the full command line used to create the ram disk?

                     

                    Regards,

                    Cristian.

                    • 52. Re: Would a RAM disk improve Server performance?
                      Toby Erkson

                      C,

                      I used the GUI interface:

                      I set the Device type as "Harddisk volume" and Image file access as "Create virtual disk in physical memory".  I did format the "drive" before use.  I was using 6GB but the test extract I'm using caused the F: drive to run out of disk space which is why I bumped it up to 10GB.

                      • 53. Re: Would a RAM disk improve Server performance?
                        Cristian Vasile

                        Toby,

                         

                        ImDisk is designed to be a small, simple and extremely lightweight driver primarily for mounting image files as virtual disks and creating ram disks with very basic features. It does not interact with certain system services like Volume Mount Manager and similar and it also lacks support for other special requests sent from services such as Resource Monitor.

                         

                         

                        Regards,

                        Cristian.

                        • 54. Re: Would a RAM disk improve Server performance?
                          Toby Erkson

                          I ran a decent sized workbook that is just over 1GB in size with multiple extracts and a custom SQL query.

                          No RAM disk:  8551 seconds

                          With RAM disk:  8237 seconds

                          Time savings:  4%

                           

                          Given the performance of other extracts and processes running on the Server I'm going to call performance increases with a RAM disk for a Tableau Server as...

                          • 55. Re: Would a RAM disk improve Server performance?
                            Cristian Vasile

                                                                                               ~~~~ The End  ~~~~

                            • 56. Re: Would a RAM disk improve Server performance?
                              Jeff Strauss

                              Toby, thank you for doing all the POC testing and sharing your results!!!  It's highly valued to not have to travel the same path / re-invent the wheel.

                              • 57. Re: Would a RAM disk improve Server performance?
                                Jeff Strauss

                                If there were such thing as designation of gold stars, you would get it...

                                • 58. Re: Would a RAM disk improve Server performance?
                                  Toby Erkson

                                  Thanks Jeff.  My testing wasn't exhaustive and it was simple, mostly because I'm not that experienced in hardware/network server stuff.  There may be tweaks that can be done that could make a difference but those would come from those of you who are more knowledgeable than I, e.g. Cristian's suggesting of also using SuperCache.

                                   

                                  I think it's important to be able to monitor the RAM disk so maybe using another application that allows it to be recognized by the system should be used instead of ImDisk.  It's a super-easy install and uninstall and seems to work fine otherwise.  Oh, and it's free & and fully-functional, something I prefer when testing.

                                   

                                  I don't think this should spell the end of a RAM disk as there's room for more experimentation.  I think the McDonald's approach I took (fast, easy, minimal brain involvement) didn't work.  However, the next iteration of the experiment may take a little more thought to get it to work.  Even if it does work it would bring up a new set of questions, like how does an admin determine the right size (my 1GB workbook consumed over 6GB during its full refresh and that was just for a single workbook running), how to get workbook downloads to function, etc.

                                  • 59. Re: Would a RAM disk improve Server performance?
                                    Matt Coles

                                    Regarding picking the right size: I set up a scheduled task on all our Server hosts to calculate the total space consumed in the temp directory for Tableau Server, every three minutes, and let it run for a week. Over that period of time, the maximum space consumed was 26,013 MB on one of our dedicated Backgrounder hosts, at peak extract refresh time around 3AM.

                                     

                                    So assuming we set up a RAM disk on each of the hosts, pointed Server at it for temp, and did not adjust the retention settings, we'd need about 35 GB of RAM for me to feel even close to comfortable in the short term. But that still leaves big questions around longer-term scaling--in order to scale up in that configuration, I'd either need to distribute the extract refresh tasks among more schedules, or cap the number we run at the same time, or (harder) the size of extracts we run concurrently. Even then, unforeseen growth could grind us to a halt, and if worst came to worst, I'd be stuck trying to order more RAM in a hurry and get it installed just to keep our extract refreshes going.

                                     

                                    None of that is really appealing to me as an admin unless we see massive performance improvements from this idea. I'm still doubtful we can achieve that with this hack, but I'm still planning on running a few experiments to test it out a bit more.