ALICE Site Requirements#
This document describes ALICE-specific requirements for sites that envisage joining the ALICE grid infrastructure at Tier-3 or Tier-2 level, in particular without custodial storage responsibilities.
Each ALICE site needs to run an "edge service" called VO-Box, which stands for Virtual Organization Box, in this case dedicated to the ALICE VO, on which ALICE-specific services (provided via CVMFS) have these responsibilities:
- job submission to the site's Computing Element(s) or batch system;
- monitoring through MonALISA of:
- jobs;
- the site's Storage Element(s);
- the site's network connectivity to other ALICE sites and to the ALICE central services at CERN.
The site maintains its VO-Box(es), while the ALICE-specific services run under an unprivileged account.
There also are requirements and recommendations for the selection and configuration of computing, storage and network services, which are further detailed below.
Services#
VO-Box#
Please consult these links:
- Generic requirements
- If a container is preferred:
- Else, assuming job submissions go through a CE:
Computing - job management#
- Jobs can be submitted directly to the batch system, if desirable
- Used at a few sites
- Most sites require use of a Computing Element
- HTCondor CE - used by most of the big sites
- ARC CE - we do not use its data handling features
- Batch systems
- HTCondor - the most popular
- Slurm
- PBS/Torque - is becoming rare at our sites
- Batch queue configuration
- Allow a maximum job lifetime (TTL) of several days, if possible
- Allow whole-node jobs, if possible
- Else 16- or 8-core job slots
- If a CE is used, allow 1-core job slots for availability tests
Computing - worker nodes#
- OS: RHEL / AlmaLinux / Rocky Linux v9
- Other Linux flavors might work, but would need to be discussed
- 50 GB CVMFS cache
- At the very least 2 GB RAM per core, preferably at least 3
- 5 GB swap per core (can also be included in the RAM)
- 10 GB job directory scratch space per core
- LAN speed: 5 MB/s per core
Storage#
- Preferably EOS or plain XRootD
- Both can be fully integrated in MonALISA monitoring
- 1 TB per core in the batch share for ALICE
- LAN speed: 5 MB/s per core in the batch share for ALICE
External network#
- Minimum speed 10 Gbps
- Larger sites: 1 Mbps per core in the batch share for ALICE
- Connection to LHCONE desirable