FCPPL computing meeting

Meeting to follow up projects initiated between IHEP Computing Center and CC-IN2P3 To connect: ------------- CONFERENCE ----------------- -- Titre : FCPPL_Vision_IV_73175 -- Date : 12 mars 2012 -- Heure : 10:00:00 heure locale -- Duree : 03:00:00 -- Numero a appeler : ---- IP : ---- ISDN : +33 (0)4 26 68 73 02 ---- TEL : +33 (0)4 26 68 73 02 -- Identifiant numerique : 73175 (terminer par #) -- Code PIN : 8372 (terminer par #) -------------------------------------------
  • Chen Gang
  • Dominique Boutigny
  • Fabio Hernandez
  • Fazhi QI
  • Ghita Rahal
  • Jerome Bernier
  • Laurent Caillat-Vallet
  • shan zeng
  • Yonny CARDENAS

Connectivity between CC-IN2P3 and IHEP

One presentation and two text files have been appended to this induce. They highlight the progress and the checks performed on IHEP and CC-IN2P3 side.


- Thorough checkups through the whole route of the transfers between IHEP and CC-IN2P3.  Tests have been performed per link segment

- Machines to perform perfsonar test are set up on various endpoints: CERNet, IHEP, CC-IN2P3.



CC to IHEP low transfer rates:


-No bottleneck was observed from London to IHEP. this was checked by Fazhi. Bottleneck likely to be on CC-IN2P3 side. 

To trace the problem, checks have been made on various machines and various subnets. It was observed that the  machines with 2 networks cards (2*1Gb/s) like ccrdli001 have a low throughput. After the meeting, Jérôme has checked that this could be due the type of configuration of the network cards (bad aggregation of the output of each card). He has found a configuration that improves the throughput (more informations?)


Another bottleneck showed up for machines with 10 Gb/s network card (like cclhcone01): they exhibit low throughput for transfers to IHEP while there is no problem for transfers to ICEPP(JP). This could be on any part of the route where the bandwidth goes from 10 to 1 or 2.5 Gb/s 


FTS parameters for CC to IHEP transfers:


CC => BEIJING: number of files=60

                       number of streams = 16

BEIJING => CC: number of files = 60

                       number of streams = 10





     - Provide CC-IN2P3 with results on the route London to CERNet (The data has already been sent to CC network experts after the meeting)

     - check of the configuration of the transfer machines at ccin2p3 (ongoing, see above). Yvan will dedicate some machines from the dcache pools to apply the new configuration of Jérôme. Checks will be done to see if this improves the throughput. If it is the case, all dcache machines will undergo a configuration change at the next CC shutdown.

     - check on the optimal FTS transfer parameters for file transfer, for example with ATLAS showcase.


Rehearsal of the presentation at the FCPPL workshop in Saclay

- Fabio will be presenting the work made in the FCPPL framework.   

There are minutes attached to this event. Show them.