[Swift-devel] VDS1 transfer executable

Mihael Hategan hategan at mcs.anl.gov
Tue Jul 24 09:37:19 CDT 2007


On Mon, 2007-07-23 at 10:40 -0500, Mike Wilde wrote:
> Im in favor of building on transfer.  There is also the newer utility "t2".
> 
> Jens Im sure will be delighted to expound on these.

Well, he said he'd use the Java stuff, since it has more flexibility
than the command line interface of globus-url-copy, which is used by
"transfer". On the other hand, the Java stuff is heavier on resources
(unless there's some form of JVM running on some form of head node
permanently).

>   As I recall one was better 
> at some things and the other at others.
> 
> For example, I think t2 will retry failing I/Os from am alternate PFN if several 
>   replicas are available.  But perhaps transfer does parallel transfers better, 
> or some such advantage.  I need to dive into old email to find the info, but in 
> the meantime Jens or the manpages can probably explain much.
> 
> - Mike
> 
> Ben Clifford wrote, On 7/23/2007 10:29 AM:
> > none of those seem to be arguments for or against rewriting vs reusing.
> > 
> > On Mon, 23 Jul 2007, Mihael Hategan wrote:
> > 
> >> Support, throttling, concurrency control. We seem to be fundamentally
> >> changing the way things work, and we do that because we can.
> >>
> >> On Mon, 2007-07-23 at 15:19 +0000, Ben Clifford wrote:
> >>> Given that the VDS1 transfer executable exists and appears to work, there 
> >>> would need to be some strong argument to not use that as a base (which 
> >>> there may be, but I don't know of one).
> >>>
> >>> On Mon, 23 Jul 2007, Mihael Hategan wrote:
> >>>
> >>>> I think the reimplementation argument is not universally valid. One must
> >>>> consider costs vs. benefits.
> >>>>
> >>>> On Mon, 2007-07-23 at 15:08 +0000, Ben Clifford wrote:
> >>>>> VDS1 has a utility, transfer, which is for use on the worker nodes to 
> >>>>> stage data in and out.
> >>>>>
> >>>>> It seems fairly seriously worth considering using that, rather than 
> >>>>> re-implementing stuff from ground up.
> >>>>>
> >>>>
> >>
> > _______________________________________________
> > Swift-devel mailing list
> > Swift-devel at ci.uchicago.edu
> > http://mail.ci.uchicago.edu/mailman/listinfo/swift-devel
> > 
> > 
> 




More information about the Swift-devel mailing list