On Thu, Jan 28, 2016 at 1:10 PM, Andy Seaborne <a...@apache.org> wrote:
> On 28/01/16 16:52, Matthew Holt wrote: > >> Hi Andy, >> Thanks for the reply. Responding in-line below... >> >> On Thu, Jan 28, 2016 at 4:26 AM, Andy Seaborne <a...@apache.org> wrote: >> >> On 27/01/16 21:45, Matthew Holt wrote: >>> >>> Hi, >>>> Wanted to check if anyone can confirm if you can successfully change the >>>> TDB File Mode in 2.7.11 >>>> >>>> >>> 2.7.1? >>> >> >> >> That's correct, I had that listed incorrectly there. >> >> >>> >>> from direct to mapped (assuming a restart takes >>> >>>> place) , or vice-versa, without any issues? >>>> >>>> >>> You can do that. >>> >>> (though upgrading would be a good idea) >>> >>> We've read that this is supported, and tried this without any issues, but >>> >>>> received some push back as at one time this was an issue. >>>> >>>> >>> What was the issue? >>> >> >> >> At this point not too sure. They encountered this issue around 5 years >> ago, >> > > Depends on who "they" are. > > (If it happens to be IBM ...) > > There is one issue that arises ... MS Windows does not allow memory mapped > files to be deleted without exiting the JVM. This is a long standing Java > issue for Sun/Oracle/OpenJDK JVMs. > > But direct mode vs mapped mode is reported as not much of a performance > win on Windows anyway. Thanks Andy. Good call there. :) That's one issue we were aware of, and for now, only enabling it for Linux 64 bit users. We hadn't seen any issue in our testing, but found a long-ago defect talking about that a potential issue, so had wanted to make sure we could be more confident with the switch over. Thanks again, Matt > > > Andy > > > and put code in place so that it would only use the File IO mode the >> indices were created with. We have only been using Direct, and are now >> switching to Mapped IO for systems that support it to take advantage of >> performance benefits. >> >> >> Thanks again for your help, >> Matt >> >> >> >>> >>> Andy >>> >>> >>> >>> Thanks for any help you can provide, >>>> Matt >>>> >>>> >>>> >>> >> >
