Dave Newbold wrote:
> Hi,
>
> I'm confused, as I thought that the passive option was simply passed
> through to the underlying transport - in this case, g-u-c.
Hi Dave,
You are probably right that the flag is passed onto g-u-c (I know
relatively little about srmcp). But regardless of what the flag does
the packets tell their own story - the SRM layer let's the client do all
the initiation (which is essentially server passive) whereas the
gridftp-server actively makes connections to the WN (which is server
active).
gridftp-server and g-u-c can work actively or passively /in single
stream mode/ - but when you do any kind parallel transfer it forces the
sending side into active mode. If we are transfering data from the SE
then the sending side is the gridftp-server and hence the process is
server active (which breaks our SNAT).
This is a mail from Bill Allcock one of the globus-url-copy authors:
http://www-unix.globus.org/mail_archive/discuss/2003/03/msg00130.html .
I think it confirms what I am saying, he says that it's a "limitation
of the current protocol". Elsewhere I think he gave a reason why it had
to be this way (a race condition thing).
> Can I suggest forgetting about the SRM layer for now, and seeing if a
> straight globus-url-copy works in one direction or the other?
It was kind of buried at the bottom of my last mail but srmcp already
works in both directions. The only caveat is that you have to use
single stream mode if you are transfering data from the SE to the WN.
regards
Jon
|