HBP/EGI Collaboration meeting

Tuesday 16 Feb 2016, 10:30 → 11:00 Europe/Amsterdam

Webex

Matthew Viljoen (EGI.eu)

Present - Lukasz Dutka (LD), Catherine Zwahlen (CZ), Christian Bernardt
(CB), Oliver Amblet (OA), Matthew Viljoen (MV)

Apologies - Björn Hagemeier

- Both CB and LD reported stability in their data hosting services over the last two weeks.  MV reiterated the importance of documenting the setup, referring to Jeff's warning at the last meeting over the complex dependencies and risks of instability.  Also, one of the outputs of the pilot work is a full documentation of the setup - this was mentioned at the HBP/EGI management meeting on 2 Feb (see separate minutes)

- CB said that his testing had demonstrated the reliable serving of 4 concurrent users for both OneData and dCache ATLASViewers.  MV said that we should aim to see how many users we can reliably service using the ATLASViewer with 'acceptable' performance - this could translate to 8 tiles being delivered under a second by the load balanced imaging services. As part of the documentation of the pilot we should predict how many resources would be required to serivce 100 concurrent users via extrapolating from our testing results.

-> Obtain maximum number of users we can service with the pilot by the end by producing timing measurements from our testing software.  By end of Feb (LD, CB)

- CZ said that a number of datastacks had been processed and were now ready to be transferred to EGI.  The biggest datastack is 2TB.  EGI should request that the INFN-Bari site has sufficient capacity to accept this data, and install the public key of the HBP person who will transfer the data.  OneData and dCache hosting sites should ensure they have sufficient capacity to host this extra data.  CZ requested that the data should have the same directory structure as HBP are used to; if the data is delivered as a tar then this requirement should be fulfilled.

->INFN-Bari to increase their space by 4TB to accept the new data stacks (MV)
->Public key to be sent to EGI of the person transferring the new datastacks (CZ)
->OneData and dCache should ensure that they have enough space to host the new datastacks (LD, CB)

- CZ said that it had been decided that the ATLASViewer be hosted by HBP rather than by EGI on the fedcloud when the service moves to production.  

-Regarding AAI, OA said that OpenID Connect is still on track to be ready for user authentication integration into the pilot by the end of Feb, and that the sample client could already be integrated into the software.  However, since ATLASViewer won't be hosted by EGI, it was recognised that it is out of scope of the pilot for EGI to integrate AAI into ATLASViewer.  Rather, it should be integrated into the imaging service.  How this should be done technically would be discussed within HBP (the imaging service software is written by HBP)

->Decide how OpenID Connect may be integrated into the imaging service (OA, CZ)

DONM: 1 Mar at 10:30 CET