Weighing in a bit late on this.
> Another issue with a central UI service is that it's more of a problem
> for users to send their jobs and also recover their output - it would
> all have to go through the central service which would be a bit of a
> pain, certainly for production level stuff.
>
> Out of interest, would it be possible for us to provide a version on
> CVMFS instead? Then users can just mount that and be confident it's up
> to date, etc.
>
> Just an idea :)
Ta da:
https://ggus.eu/ws/ticket_info.php?ticket=96030
There is a tarball UI in cvmfs, although it's only set up for the big
three + dteam. You need to enable the grid.cern.ch repo to have a gander
at it. There's worker nodes in that there repo too.
I'm interested in what Steve was proposing, as I can easily roll that
into the tarballs - making them much more complete.
Cheers,
Matt
> THanks,
>
> Mark
>
> On 29/08/13 23:44, Martin Bly wrote:
>> The problem with the central UI service at the Tier1 was not the UI
>> systems in particular, since we had (and still have) them anyway - it
>> was (would be) managing and keeping track of all the remote users.
>> Particularly when they go dormant / disappear without telling you.
>>
>> Martin
>> --
>> Tier1 Fabric Team Leader
>> ________________________________
>> From: Ewan MacMahon<mailto:[log in to unmask]>
>> Sent: ý29/ý08/ý2013 15:39
>> To: [log in to unmask]<mailto:[log in to unmask]>
>> Subject: Re: Instant UI
>>
>>> -----Original Message-----
>>> From: Stephen Jones [mailto:[log in to unmask]]
>>> Sent: 29 August 2013 15:26
>>>
>>> On 08/29/2013 03:05 PM, Ewan MacMahon wrote:
>>>>> -----Original Message-----
>>>>> From: Testbed Support for GridPP member institutes [mailto:TB-
>>>>> [log in to unmask]] On Behalf Of Stephen Jones
>>>>>
>>>>> With the working title "Instant UI", the aim of the project is to
>>> produce
>>>>> a small suite of documentation and software comments that will
>>>>> enable a
>>>>> new user to set up a UI and join the grid with the minimum of hassle.
>>>> Much as I dislike being negative[1]; is this actually a good idea?
>>>> If the question is how to get on the grid with minimal hassle, then
>>>> the answer is surely not 'install a UI' - the answer is 'find someone
>>>> with a UI and get them to give you a login'.
>>> While that is undoubtedly easy, unfortunately I have at least one new
>>> user who hasn't got a UI in his entire institute. I guess he could ssh
>>> to one, which (say) we maintain for him. His line of business is "drug
>>> discovery" in the education sector, i.e. drug site binding algorithms
>>> etc. His students could benefit if they could demonstrate knowledge and
>>> use of grid computing etc. to find novel molecules. It could be an
>>> important
>>> area. Neasan (before he left) requested new users and uses, and we've
>>> got
>>> one. But now he's gone and I'm stuck - I have to offer something.
>>>
>> Sure, it sounds completely worth supporting, but offer him a UI; it
>> should be considerably less time and effort for you to even create a
>> brand new one on a VM than it would be to document and support him
>> through creating his own. Call it a VO box, and you're done.
>>
>>>> While the UI may be one of the simpler grid service nodes, they're
>>>> not trivial, they do require active maintenance (preferably by
>>>> someone who receives EGI broadcasts), and they have a habit of failing
>>>> in spectacularly odd manners if they don't get it.
>>> Thanks; that's a new requirement for Instant UI (should it come
>>> to fruition) - i.e. it needs maintenance procedures etc. It grows and
>>> grows!
>>>
>> Like the plague.
>>
>>>> if we're serious about supporting 'other' VOs rather than just
>>>> being a backend for the WLCG, then we don't need to put the effort
>>>> into documentation, we should put it into running a central UI again.
>>> That's certainly a route to consider. It's not a case of
>>> "either/or" - we could have both or (as now) continue
>>> with neither. Let's have a think - who would fund and implement
>>> the "central UI"? It's a good idea, but that's the problem.
>>>
>> Well, since we're talking about requirements for GridPP5, I
>> would suggest in outline a restoration of the Tier 1 UI service,
>> funded out of the same 'other VOs' allocation of the money that
>> funds stuff like the WMSes and the LFC etc. And if there isn't
>> any money for supporting 'other VOs', then we don't support them
>> and the need for a UI disappears.
>>
>> Ewan
|