Andrew Gallatin wrote:
Rao Shoaib writes:
> >
> Sure I am ccing NFS discuss so that they can look into this again if
> they are not already. It will be interesting to know if there is extra
I'm not on that list, so my mail to them will likely bounce.. :(
Well you should be you have some good ideas to share :-).
> over head in doing user to kernel copy than just kernel to kernel copy
My understanding is that copyin is just bcopy with an on-fault handler
to handle any potential page faults rather than taking a kernel trap.
So, for a bulk data copy where all pages are present, it is
essentially bcopy.
Eric will have to verify this. I am not sure if there will any over head
in accessing the user address space as compared to kernel address space
and if there will be any more or less page faults.
> (which will be the FS case). BTW as you may know sendfile does not
> always uses zero copy. I am assuming in your measurements it was using
> zero copy.
Yes. I spent quite a while cursing while I hacked on my test program
to make sendfile do zero-copy. Thank goodness Solaris is open
source, else it might have taken me much longer to figure out.
Yes It's so easy to make sure that your application works properly when
you have access to the kernel code.
Rao.
Drew
_______________________________________________
networking-discuss mailing list
[email protected]
