fvdw wrote:See here
That is about the cifs kernel module, which implements the client filesystem driver. (Needed for 'mount -t cifs //server/share /mountpoint). There is also a server kenel module,
ksmbd, but the package which is called 'samba' or 'samba-server' on most distro's uses neither of them, and is pure user space software.
Besides that it seems that WD also saw the performance issue and introduced a kernel using 64k pages to improve transfer speeds. Unfortunate our firmware doesn't work with such a kernel.Maybe this 64k based kernel is much more effectuve to improve transfer speeds than the smb protocol.
I doubt that is was used to speed up Samba. The ZyXEL NAS540 originally also used 64k pages, but when they changed that to 4k the samba speed only slightly
increased.
I think the main reason that ZyXEL originally used 64k pages was the filesystem size. Ext3 can only have 2^32 clusters, and a cluster has to fit in a page. (Don't know why exactly). And while for ext4 the number of clusters can be bigger, that was not yet the case for the stock kernel. And so with 4k pages the filesystem size was limited to 16TiB (-4k), which can be restricting for a 4 bay NAS. With 64k pages the filesystem size could be 256TiB.
Maybe WD used 64k for the same reason. AFAIK there is a 2 disk version of the My Cloud, which could exceed the 16TiB limit with two >8TB disks in raid0.
(BTW, the use of a sector size >4k made the volumes hard to mount on other boxes. The ext3/4 kernel driver doesn't accept that. But there is a work-around, there exists an ext3 fuse implementation (for BSD and Mac, I suppose) which doesn't care about page size.)