"Edward \"Koko\" Konetzko" <konetzed@xxxxxxxxxxxxxxxxx> wrote:
> > If you search through the patch, you'll find two lines with 'kdebug' on
> > them. If you turn the kdebug to _debug or just comment out the lines and
> > retry.
> >
> I have changed kdebug to _debug and tested, issue seems to be gone. NFS now
> seems really slow to be though,
If you could turn:
+ _debug("discard tail %llx", oi_size);
back to:
+ kdebug("discard tail %llx", oi_size);
I'd be interested in knowing if that's triggering at all. If it is, that
could be slowing things down (it indicates a truncation will be performed).
In your statement above, are you comparing NFS with a cache to NFS without? If
so, that might well be true: the cache has to eat some of your client machine's
performance in order to give something back, so it is not an automatic win.
If your machine is, say, connected by GigE to the server, and the server has
your working set of files in RAM over your tests, then going to the server's
RAM will very likely be faster than going to local disk.
Where caching helps is with slow connections and highly contended networks, and
in the latter case, it may not speed the client up appreciably, but it may
massively lower the network loading.
David
--
Linux-cachefs mailing list
Linux-cachefs@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/linux-cachefs
