ALICE visit

== Patricia presented a detailed description of ALICE services ==

Questions resolved, action items.

 - share software are can be mounte r/o on the worker nodes. It only needs to be writable by cmssgm on the VOBOX

 - vo box proxy. ALICE doesn't need to read the provate key of the host certificate.

 - log files: ALICE is working on syslog plugin in services. Meanwhile, log file location is determined and log files can be harvested by a site for required 90 day acrchiving. Action item: Manfred asked to update the VOBOX questionaire to provide location of the log files of AliEn services. 

== Xrootd - Fabrizio ==

Jobs are always sent to where the data is. Global redirector helps where a file is for any reason missing from the site at job runtime. Thus in fact it helps to fill 'holes' in the site storage. Such 'haotic' inter-site traffic is pretty negligent. 

ALICE recognise the problem of protecting data on tape from general user access. Trivial soluton is to change permissons on the AliEn catalog, then only authorized users will be able to ask for such file. Part two is education. Part three is that it will not work well for users, and thus users will naturally use AODs in their jobs instead.

== Access to storage - Latchezar ==

Latchezar presented simple pictorial view of data movement in ALICE

GridKa is ~50% of ALICE T1s

ALICE recognises that recall from tape is an open issue, both at CERN and sites. Thus, ideally would try to avoid recalls and have disk in front of tape as large as data volume on tape. Lathezar suggests that disk in front of tape should be 80% of total disk capacity of ALICE, while 20% is pure disk-only space, no backup on tape.

Conditions data servers - all at CERN. 

== xrootd at Gridka - Artem ==

In addition to the existing disk-only xrootd system, Artem proposed to setup an xrootd cluster with tape backed and BestMan/SRM front-end.
 This solution will satisfy Alice requirements for custodial data storage and access. ALICE supported this proposal. It was decied to setup such new cluster in parallel to existing dcache tape-backed SE. Alice will be able to replicate data thus avoiding a risk of data loss. Alice will be able to swich RAW data flow back to the exising dCache instance in case the new one is not performing well.