You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
While 8192 b seems like a reasonable default for reading files, on modern systems it might be a bit too small, especially for sequential access. We should consider bumping it up a bit.
Yea, good point! I picked 8kb as its a nice default for packet size in networking. I think there's perhaps two things we could do, maybe one or the other, or both:
Have a configuration option for FsPool (or just read?) on preferred buffer/read size.
Check for st_blksize on cfg(unix) before starting to read a file.
Perhaps the best is to have the config option, and prefer that over st_blksize if set, so people can still explicitly say things like "even though my file system likes 128kb buffers, I don't want more than 32kb of memory used at a time". What do you think?
While
8192 b
seems like a reasonable default for reading files, on modern systems it might be a bit too small, especially for sequential access. We should consider bumping it up a bit.I ran a couple of tests on my NAS:
Note that I'm using ZFS, which doesn't support
sendfile()
. The128 kb
value is the filesystem preferred buffer size (st_blksize
instruct stat
).EDIT: notice how
hyper
withfutures-fs
was faster thannginx
👍.The text was updated successfully, but these errors were encountered: