On 08/02/2013 09:47, Andrew Lahiff wrote:
> Hi,
>
> CMS does not require separate queues. Sites are free to use SL5 or SL6 worker nodes, or some of each. However, sites intending to move to SL6 need to ask the glideinWMS ops people to set GLIDEIN_REQUIRED_OS to "any", because the pilots check the OS installed on the WNs and by default expect it to be SL5.
Ian Fisk didn't give this detail yesterday. He did say the code is
incompatible in both ways and code that is compiled on either OS is not
compatible with the other.
He also made a point to leave lxplus on SL5 until most grid resources
are on SL5.
cheers
alessandra
> Regards,
> Andrew.
>
>
> -----Original Message-----
> From: Testbed Support for GridPP member institutes [mailto:[log in to unmask]] On Behalf Of Alessandra Forti
> Sent: 07 February 2013 15:38
> To: [log in to unmask]
> Subject: Re: SL6 WNs - when would sites want to move?
>
> It depends on the experiment. CMS has different executables and needs at the very least separate queues. Atlas uses compatibility libraries and can run SL5 releases on SL6 so it is a bit smoother but they are not ready to go on a big scale. Lhcb are like CMS I think.
>
> All experiments agreed that we can talk again after Moriond.
>
> cheers
> alessandra
>
>
> On 07/02/2013 15:27, Stephen Jones wrote:
>> On 02/07/2013 03:05 PM, Jeremy Coles wrote:
>>> Dear All,
>>>
>>> I would like to understand whether any GridPP sites have a strong
>>> need to move their WNs to SL6 and what would the sysadmins on this
>>> list consider as a reasonable timeline on which to migrate their WNs
>>> to SL6?
>>>
>> Hi Jeremy,
>>
>> We (Liv) don't "need" to move yet. There are some technical issues
>> surrounding the migration, and we might need to do a bit of research.
>>
>> Can a cluster be mixed sl5 and sl6 and run the same jobs, i.e.
>> is it a big bang or an incremental migration?
>>
>> If it's an incremental migration, we can put (say) 4 x WNs on SL6 and
>> see how it goes. It's dead easy, and no risk.
>>
>> But if it's a big bang migration, we need more coordination.
>> If we want to be able to rollback, we will need separation, via some
>> scheme (spare CEs, two batch head nodes, two queues, scheduling rules
>> or whatever). We'd make the transition to the new SL6 setup, and
>> gradually retire one set while bringing the other on line. But this
>> will need some careful thinking through to dodge the constraints!
>>
>> Does anyone already have a good plan in their back pocket? Can I see
>> it, please?
>>
>> Steve
>>
>>
>>> Many thanks,
>>> Jeremy
>>
>
> --
> Facts aren't facts if they come from the wrong people. (Paul Krugman)
--
Facts aren't facts if they come from the wrong people. (Paul Krugman)
|