[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
Re: [ptp-dev] Parallel Environment implementation for PTP Resource Manager model
|
Greg
I have a meeting that ends at 3:00, so I might be a few minutes late, but
that time works for me.
Dave
Greg Watson
<g.watson@compute
r.org> To
Sent by: Parallel Tools Platform general
ptp-dev-bounces@e developers <ptp-dev@xxxxxxxxxxx>
clipse.org cc
Subject
12/15/2006 02:07 Re: [ptp-dev] Parallel Environment
PM implementation for PTP Resource
Manager model
Please respond to
Parallel Tools
Platform general
developers
<ptp-dev@eclipse.
org>
We can use my conference call number (the one we use for the PTP
meetings). I'm pretty free most of next week. How about 3pm EST/1pm
MST on Monday 18th?
Greg
On Dec 15, 2006, at 11:14 AM, Randy M. Roberts wrote:
> Dave,
>
> It's great to have you on board at this time. We are currently trying
> to hash out ideas on these subjects into resource manager requirements
> and a design. Issue 2) is especially vexing, and we could do with
> as much input as possible.
>
> I'll be here next week, but I don't know anything about setting up
> conference calls.
>
> Regards,
> Randy
>
> On Fri, 2006-12-15 at 10:24 -0500, Dave Wootton wrote:
>> I have my protoype code that implements PE support working, I think
>> correctly with the PTP head level code I grabbed Tuesday night.
>> The job
>> view seems to update correctly, my PE implementation shows up as a
>> resource
>> manager that I can start and stop, and when I invoke an
>> application using
>> the PE resource manager, I can see the machine view update with state
>> changes from available to running to completed as my job runs. It
>> isn't
>> perfect, but at least I'm somewhere in the right area.
>>
>> So, if you have the time later next week, I'd like to have a
>> conference
>> call to discuss a couple things that I've been trying to resolve, and
>> whatever else may come up.
>>
>> 1) How should the machine view deal with how PE can be used
>> interactively,
>> where there is no global daemon running to trach the state of the
>> cluster,
>> and all a PE process knows about is itself? I can submit multiple
>> PE jobs,
>> each, in principle, with their own host list and therefore
>> different set of
>> nodes. This implies that the machine view can add and delete
>> nodes. This
>> also raises questions about deleting nodes, since otehr things may
>> care
>> about node numbering.
>>
>> 2) What's the direction for specifying things like PE environment
>> variables
>> to run an application? The simplest solution is to just specify a
>> script
>> file that contains the environment variables, and in essence, run
>> that
>> before running the application. This puts the burden on the user
>> to set the
>> script up. The other extreme is to have settings panels, either in
>> preferences or launch configuration, that provide checkboxes,
>> buttons,
>> droptdowns, etc, and maybe tooltip text so a new user sees
>> something a
>> little more friendly, but which puts the burden of managing the
>> compexity
>> of environment variable rules on the maintainers of the PTP code.
>> Maybe
>> both are required, to support both the new and experienced users.
>>
>> I don't have any way to set up a conference call now (I should
>> work on
>> that), so if we can do this, I'll need somebody else to set one up.
>> Somebody mentioned Thursday being better. That works for me, and I am
>> around all week. If you would rather wait until January, I can do
>> that too.
>>
>> Thanks
>> Dave
>>
>> _______________________________________________
>> ptp-dev mailing list
>> ptp-dev@xxxxxxxxxxx
>> https://dev.eclipse.org/mailman/listinfo/ptp-dev
>
> _______________________________________________
> ptp-dev mailing list
> ptp-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/ptp-dev
_______________________________________________
ptp-dev mailing list
ptp-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-dev