[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
Re: [geclipse-dev] Some problem when implement interface on anothermiddleware
|
Dear Yuanbin Zou,
for debugging we don't offer to send jobs via a resource broker, but to
a computing element (CE) directly.
Therefore you need to specify the hostname of this CE and its jobmanager
(i.e. hostname/jobmanager-pbs but preferably hostname/jobmanager-fork
because its faster).
Instead of submitting the application itself as a job, we run glogin as
a job on the CE or its respective workernodes.
Glogin connects back to g-Eclipse and offers an interactive connection
similar to ssh which is used for debugging and remote building.
We currently use the globus 2 version of glogin (glite supports globus 2
jobs) however with i2glogin, a version which acts as an interactive
agent, is available.
glogin url: http://www.gup.uni-linz.ac.at/glogin/
There are several papers available dealing with grid debugging or
interactive connections:
"Net-dbx-G: a Web-based debugger of MPI programs over Grid
environments"
"Worqbench: An Integrated Framework for e-Science Application
Development"
"A WSRF-Compliant Debugger for Grid Applications"
"A debugger for computational grid applications"
"glogin - A Multifunctional, Interactive Tunnel into the Grid"
"Traffic Forwarding with GSH/GLOGIN"
Best regards,
Christof
On Thu, 2008-03-27 at 16:36 +0800, 邹园斌 wrote:
> Dear Mathias,
>
> Thanks very much for your reply. That is just what I want to ask and
> your answer is so helpful to me.
> So, the situation is the same in gLite and the middleware I am going
> to handle.
>
> I planned to develop some agent on the grid server to run some debuger
> and contact with the outsiders which was supposed to be the g-Eclipse.
> Then the work would turn to be contacting with this agent.
>
> So, I want to know how the gLite handle this problem detailedly. Is it
> using some agent like what I mentioned or some other solution. If so,
> is this agent a part of the middleware or developed as some service by
> grid service developer. I know little about the debuger and debuging
> procedure of the MPI, so I really appreciate if you can give some
> reply.
>
> Best wishes,
> Yuanbin Zou
> FIT 1-111, Grid Computation Department, DSCT, Tsinghua University
> Beijing, 100086
>
> 2008/3/27, Stuempert, Mathias IWR <mathias.stuempert@xxxxxxxxxx>:
> Hi Zou,
>
>
>
> I'm not quite sure if I got your question right. I think you
> are asking how we are able to remotely compile or debug
> applications running on the Grid, taking the border condition
> of non-interactivity of batch systems into account, right?!
> Well, in fact we are not sending jobs to a queue for compiling
> or debugging purposes but directly work with the worker nodes
> themselves, i.e. we use gLogin to access the worker nodes
> and/or computing elements and to communicate with them in
> order to be able to exchange executables and/or debug
> information.
>
>
>
> But to be honest I am not the expert on that. Nevertheless I
> am sure Thomas and Christof (also on this mailing list) can
> give you more details.
>
>
>
> Best regards,
>
>
>
> Mathias
>
>
>
>
> ______________________________________________________________
> Von:geclipse-dev-bounces@xxxxxxxxxxx
> [mailto:geclipse-dev-bounces@xxxxxxxxxxx] Im Auftrag von ???
> Gesendet: Donnerstag, 27. März 2008 05:05
> An: Developer mailing list
> Betreff: [geclipse-dev] Some problem when implement interface
> on anothermiddleware
>
>
>
>
> As I know, almost all jobs in Grid are managed by batch
> systems such as PBS or LSF. These batch systems run the jobs
> in a queue, always not instantiously and certainly without any
> interactive I/O. While users use the Grid environment to debug
> their projects, such as MPI projects, these properties are
> intolerable. So I am wondering how gLite and g-Eclipse handle
> this problem. I have noticed that there is such usage in the
> help document.
>
> Best wishes,
> Yuanbin Zou
> FIT 1-111, Grid Computation Department, DSCT, Tsinghua
> University
> Beijing, 100086
>
>
>
> _______________________________________________
> geclipse-dev mailing list
> geclipse-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/geclipse-dev
>
>
> _______________________________________________
> geclipse-dev mailing list
> geclipse-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/geclipse-dev