You are not logged in.
Hello,
I would just like your opinions and maybe experiences with this...
I have a PC with tons of free RAM and also a fast SSD. Although tmpfs (therefore RAM) is obviously faster than any SSD, do you think using anything-sync-daemon for frequently accessed files would make a noticable difference in terms of user experience? Is it worth it?
I am asking this also becase when you have enough RAM, the kernel will cache frequently accessed files to RAM anyway, so using something like anything-sync-daemon or profile-sync-daemon could possibly even be an overhead...
Thanks for your thoughts on this.
Regards,
Davor
Offline
Both can use an overlayfs mount which is not overhaad. Both are spdesigned to avoid writes and use faster ram as you pointed out.
Offline
Thank you for your response.
You are right that write performance should be superior because actual writes to disk are delayed. What about reads considering that those files are probably already in RAM cache, kernel does that caching by default. For an example, look at the column 'cached" in the output of the command "free". That is the amount of filesystem data and metadata that is already in RAM. And regarding writes, i already have all filesystems mounted with options "noatime,commit=600". The first one assures that no metadata is written to disk if files are only accessed, for example read from. The other one instructs the VFS to write files to disk only every 10 minutes, of course if there is enough memory and unless the application opens the file in O_DIRECT mode or issues explicit fsync call or if you force the sync with the "sync" command. All well behaving applications do that when they absolutely have to. So I'm not sure a about possible benefits even in writes when using anything-sync-daemon in my case considering vast amounts of free RAM I have and the configuration I described above.
Thanks,
Davor
Offline
commit=600 seems excessive and dangerous to me but I am not that familiar with all the caveats of using it.
Offline
I had that config for years, it is a bit excessive but I didn't have any problems with it, you just have to be aware you have that option. As I mentioned, applications that absolutely have to write to disk immediately do so explicitly by issuing fsync or open the files with O_DIRECT. If you umount the filesystems cleanly, fsync is also issued automatically. This happens for an example during a clean shutdown. The problem would be pulling the plug or such, but nothing irreparable, probably losing a few files. I did that a couple of times and never ended with filesystem corruption. What I do is that I issue manually the "sync" command just in case after doing some serious filesystem activity, for an example after "pacman - Syu" or after copying some big file or many files to disk. Also before shutdown or reboot, but that is really only a precaution if shutdown stucks for some reason before umounting the filesystems, but that doesn't really happen.
Thanks,
Davor
Offline