> > Why does data-only sync exist? Is it a perf thing? Reliability? Other?
Performance. On Linux/ext4 fdatasync() flushes only the file's contents instead of flushing contents + metadata as fsync() would do. For some workloads this can be a 2x improvement. Best, Thiemo On Mon, Apr 6, 2015 at 7:30 PM, Howard Kapustein < Howard.Kapustein at microsoft.com> wrote: > >data-only sync does not have a performance advantage over a full sync > Does data-only sync have a perf advantage w/synchronous=normal? > > Why does data-only sync exist? Is it a perf thing? Reliability? Other? > > > >Our experience with various Unix flavors teaches us > Windows != Unix. Although FlushFileBuffers is conceptually equivalent to > fsync we're well into the realm of details mattering. > > > >Nt*() interfaces may cause portability problems for things like WinCE > Not just CE - NtFlushBuffersFileEx was added in Win8. But SQLite can and > does use functions new in the last decade. A simple > LoadLibrary/GetProcAddress can detect availability at runtime, and there's > already plenty of #ifdef options. I already build a custom DLL from > amalgamated source (for reasons) and I'm targeting Win8+ so adding > /DSQLITE_ENABLE_WIN_DATASYNC would be trivial. Likewise enabling this > #ifdef SQLITE_OS_WINRT is safe enough since WinRT only exists on Win8+ > > > I'm trying to understand (a) if this has been considered, (b) if it's > planned, and (c) if I wanted to hack it myself, do the resident experts > have any implementation advice? Both to make it work, and if it's useful > what's likeliest to be smoothest to be accepted as a patch. > > - Howard > > -----Original Message----- > From: Joe Mistachkin [mailto:joe at mistachkin.com] > Sent: Friday, April 3, 2015 10:35 AM > To: 'General Discussion of SQLite Database' > Cc: Howard Kapustein > Subject: RE: [sqlite] NtFlushBuffersFileEx for SQLITE_SYNC_DATAONLY > onWindows ? > > > Howard Kapustein wrote: > > > > Has anyone considered supporting SQLITE_SYNC_DATAONLY on Windows using > > NtFlushBuffersFileEx? > > > > http://msdn.microsoft.com/en-us/library/windows/hardware/hh967720(v=vs.85).a > spx > > > > Our experience with various Unix flavors teaches us that data-only sync > does not have a performance advantage over a full sync. And the Nt*() > interfaces may cause portability problems for things like WinCE. > > -- > Joe Mistachkin > > _______________________________________________ > sqlite-users mailing list > sqlite-users at mailinglists.sqlite.org > http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users > -- Thiemo Nagel | Software Engineer | tnagel at google.com | +49 89 839309091
