On Tue, Feb 22, 2005 at 03:00:20PM +0000 or thereabouts, Owen Synge wrote:
> On Tue, 2005-02-22 at 14:55 +0000, Alessandra Forti wrote:
> > > Because data is transfered to and from these servers over the internet.
> >
> > So it skips head node and gridftp servers second layer (in the tier1
> > architecture) as explained by Andrew few mails ago?
>
> The process is as follows,
>
> {A} Client contacts server and asks where to put the data
> {B} Client Receive a "Transfer" URL which is where to place the data
> {C} Client uploads the data to URL.
Though in the case of SRM copy method is it?
{A} Client contacts 1st SRM and asks it to transfer file to another 2nd SRM
{B} 1st SRM asks 2nd SRM where to put the file.
{C} 1st SRM tells backend dcache nodes to put the file where the 2nd SRM
told it to put it.
Steve
>
> Regards
>
> Owen
>
>
> > On Tue, 22 Feb 2005, Owen Synge wrote:
> >
> > > On Tue, 2005-02-22 at 14:41 +0000, Alessandra Forti wrote:
> > >> So if there is no difference why using srmcp you need the data servers
> > >> exposed to the internet?
> > >
> > >
> > >> Are they SRMs themselves?
> > >
> > > No they are part of the SRM system but SRM its self is a redirecting API
> > > which returns URL's to the data server.
> > >
> > >> Pehaps I haven't
> > >> understood what an SRM is. I thought dcache was one of the SRMs, is it
> > >> not?
> > >
> > > Dcache provides an SRM API and data servers to support SRM use so I
> > > should say it is.
> > >
> > > Owen
> > >
> > >>
> > >> cheers
> > >> alessandra
> > >>
> > >> On Tue, 22 Feb 2005, Steve Traylen wrote:
> > >>
> > >>> On Tue, Feb 22, 2005 at 02:14:56PM +0000 or thereabouts, Owen Synge wrote:
> > >>>> On Tue, 2005-02-22 at 14:05 +0000, Alessandra Forti wrote:
> > >>>>> Hi Steve,
> > >>>>>
> > >>>>> as usual we are not talking the same language :) and I need a bit of
> > >>>>> explanation.
> > >>>>>
> > >>>>>> I think everything is correct. Only addition is that the difference between
> > >>>>>> CMS and the data challenge is that CMS do srmcp's where as the service
> > >>>>>> challenge does a (get-turl && get-turl) , 3rd party g-u-c,
> > >>>>>> (close-turl && close-turl). This is why the disk servers them selves
> > >>>>>> must be exposed to internet in the CMS case since it is them that do the
> > >>>>>> transfer for you and they do it non-passivly.
> > >>>>>
> > >>>>> what package does srmcp belong?
> > >>>>
> > >>>> d-cache-client-1.0-34
> > >>>>
> > >>>>> what package do get-turl/3rd party g-u-c/close-turl belong?
> > >>>>
> > >>>> Any Srm client. So probably the experimental code for CMS is my guess in
> > >>>> this case.
> > >>>>
> > >>>>> what does g-u-c mean?
> > >>>>
> > >>>> I am guessing Steve means Globus URL Copy
> > >>>>
> > >>>>> what is the difference between the two?
> > >>>>
> > >>>> srmcp is written in Java, the rest are in C/C++.
> > >>>>
> > >>>> srmcp supports dcap.
> > >>>>
> > >>>> Apart from that, some WSDL passing, and some other trivial stuff not
> > >>>> much from my knowledge, can you add more Steve?
> > >>>
> > >>> Not much, The srmcp supports the SRM cp method where you can ask one
> > >>> SRM to copy a file to another SRM on your behalf thus making the
> > >>> transfer asyrchonous.
> > >>>
> > >>> lcg-rep does a (get-source-turl,get-dest-turl), 3rd party globus-url-copy,
> > >>> (close-source-turl,close-dest-turl) and is sychronous.
> > >>>
> > >>> Steve
> > >>>
> > >>>
> > >>>
> > >>>>
> > >>>> Owen
> > >>>
> > >>> --
> > >>> Steve Traylen
> > >>> [log in to unmask]
> > >>> http://www.gridpp.ac.uk/
> > >>>
> > >>
> > >> --
> > >> ********************************************
> > >> * Dr Alessandra Forti *
> > >> * Technical Coordinator - NorthGrid Tier2 *
> > >> * http://www.hep.man.ac.uk/u/aforti *
> > >> ********************************************
> > >
> >
> > --
> > ********************************************
> > * Dr Alessandra Forti *
> > * Technical Coordinator - NorthGrid Tier2 *
> > * http://www.hep.man.ac.uk/u/aforti *
> > ********************************************
--
Steve Traylen
[log in to unmask]
http://www.gridpp.ac.uk/
|