[RU-NGI] Fwd: Multicore status and plans

[Victor Kotlyar]

Здравствуйте, Андрей Андреевич!

Очереди созданы.

ce0004.m45.ihep.su:8443/cream-pbs-atlasmc
ce0002.m45.ihep.su:8443/cream-pbs-atlasmc

Ограничение по ЦПУ не больше 8 на задачу.

С уважением
Виктор Котляр


on 07.10.2014 19:03, Andrey Minaenko wrote:
> Пересылаю предыдущий мэйл. Кодировка д.б. UTF-8. Надеюсь сейчас читабельно.
>
> Здравствуйте, уважаемые коллеги,
> АТЛАСовские production jobs полностью переведены на многоядерный режим
> (см. письмо ниже). Поэтому для того, чтобы получать соответствующие
> задания необходимо во всех АТЛАСовских сайтах создать multicore queues,
> и сообщить мне их названия. В сайтах, которые поддерживают также задания
> анализа (IHEP, JINR, RRC-KI) нужно приписать 60% АТЛАСовских узлов к
> этой новой очереди, в остальных сайтах - 100%. Подробнее - см. ссылки в
> письме ниже.
> С уважением, Андрей.
> P.S. Просьба сделать это как можно быстрее, т.к. сейчас Ru-Tier-2
> практически заданий АТЛАСа не получает. Только небольшой поток заданий
> анализа.
>
> ---------- Forwarded message ----------
> Date: Tue, 30 Sep 2014 13:44:11 +0200
> From: Alessandra Forti<Alessandra.Forti at cern.ch>
> To:"atlas-adc-operations at cern.ch" <atlas-adc-operations at cern.ch>
> Cc: Alessandro Di Girolamo<Alessandro.Di.Girolamo at cern.ch>,
> "andrej.filipcic at ijs.si >> Andrej Filipcic"<andrej.filipcic at ijs.si>,
> Simone Campana<simone.campana at cern.ch>, Torre Wenaus<wenaus at gmail.com>
> Subject: Multicore status and plans
> Dear Clouds and Sites,
>
> thank you very much to all those sites with existing resources who acted
> last
> week. The result is evident by the sharp increase of successful jobs
> throughput.
>
> ss
>
> I'd like now to turn to the sites which still don't have a multicore
> queue. The
> plan for the future is to run all central production in multicore and
> for this
> ALL sites need a multicore queue enabled. At the moment though not all
> sites
> have it. I gave a presentation of the status at the S&C week earlier in the
> month [1] with some stats on the number of queues (slide 8). Here is
> reported
> directly
>
> Leaving out T3s for now. There are
>
> * 12+84 sites listed in AGIS T1+T2
> * 12+48 production MCORE T1+T2 but they are not all active in the
> dashboard I
> count 11 T1 and 43 active T2.
>
> So it seems we have 5 T2 with a queue but they are not running and 36
> T2s are
> without an MCORE queue altogether. Without listing the sites for now can
> the
> cloud squads and the sites look into add the sites that don't have a
> queue yet
> and for those 5 with queues that are not yet running to fix them?
>
> If you are unsure about how to set up the batch system please follow the
> advice
> of the WLCG multicore TF [2]. Most of the advice of the TF is aimed at
> minimizing the problems that come with scheduling jobs requesting different
> number of cores as said in the slides in [1], but if you have a solution
> that
> you trust please go ahead and apply it.
>
> Let me know if you have any questions.
>
> thank you for your cooperation
>
> cheers
> alessandra
>
> [1]
> https://indico.cern.ch/event/276501/session/11/contribution/25/material/slides/0.pdf
>
> [2]
> https://twiki.cern.ch/twiki/bin/view/LCG/DeployMultiCore#Batch_system_related_information
>
>