Thread: ToDo: preload for fulltext dictionary
Hello, Czech users reports a slow first fulltext queries. It is based on using ispell dictionary. The dictionary data could be shared or minimally dictionary could be preloaded like some PL language. What do you think about this? Regards Pavel Stehule
Pavel Stehule <pavel.stehule@gmail.com> wrote: > The dictionary data could be shared or minimally dictionary could be > preloaded like some PL language. > > What do you think about this? Surely preloading is the most realistic approach, but I hope we would support dynamic allocation of shared memory, and load dictionaries in the area and share it with backends. We should avoid additonal calls of shmget() or mmap() in the additional shared memory allocation, but we can shrink shared buffers and reuse the area for general purposes. We often have serveral GB of shared buffers nowadays, so dividing some MB of buffers will not be problem. Regards, --- Takahiro Itagaki NTT Open Source Software Center
2010/2/16 Takahiro Itagaki <itagaki.takahiro@oss.ntt.co.jp>: > > Pavel Stehule <pavel.stehule@gmail.com> wrote: > >> The dictionary data could be shared or minimally dictionary could be >> preloaded like some PL language. >> >> What do you think about this? > > Surely preloading is the most realistic approach, but I hope we would support > dynamic allocation of shared memory, and load dictionaries in the area and > share it with backends. We should avoid additonal calls of shmget() or mmap() > in the additional shared memory allocation, but we can shrink shared buffers > and reuse the area for general purposes. We often have serveral GB of shared > buffers nowadays, so dividing some MB of buffers will not be problem. > I am for all solution, but using a shared memory isn't realistic now, and can be a problem on MS Windows. More - it can be difficult with some special custom non ro dictionaries. Regards Pavel > Regards, > --- > Takahiro Itagaki > NTT Open Source Software Center > > >