More than 250 sites from all over the world provide computing services for the particle-physics community, for example in the framework of the Worldwide LHC Computing Grid (WLCG) project. Many sites are facing the same challenges and problems, so why not share insights and solutions? That's the idea behind HEPiX, an informal organization that holds workshop-style meetings twice a year.

The spring 2009 meeting was hosted by Umeå University in Sweden on 25–29 May. Umeå is a town on the Baltic coast, only 350 km south of the Arctic Circle.

Academia is very important in Umeå: little more than 100,000 people live in the town and the two universities have almost 30,000 students. The Scandinavian countries contribute a special Tier 1 to LCG – it is distributed among a number of sites, one of which is Umeå.

Some 100 attendees registered, surpassing the expectations of the HEPiX board and the local organizing committee, led by Mattias Wadenstein. There were more than 50 scheduled presentations, organized in tracks covering virtualization, storage and file systems, operating systems and applications, security and networking, and data centres. Full details are available from Indico: http://indico.cern.ch/conferenceDisplay.py?confId=45282.

Virtualization

This track attracted the most interest. Although forward-looking presentations on the technology and vague usage ideas were still around, they were complemented by reports concerning production experience and benchmarking. This showed that virtualization has matured. For example, it is being used in production facilities to provide high-availability services, a large and highly flexible test facility for software testing, the possibility to run experiment-specific images, etc. Security concerns were also discussed. Detailed comparisons showed the strengths and weaknesses of popular hypervisors, such as XEN, KVM, VMware and HyperV, and demonstrated that under certain circumstances virtual machines can even perform better than physical ones. The track then focused on computing clouds, mentioning both commercial and academic solutions including first cost estimates and the significant security aspects. A full Grid site has already been successfully run on a commercial cloud.

Storage and file systems

Storage was a central theme. The session provided an overview of the HEPiX file system Working Group activities as well as presentations targeted at specific technologies. The Working Group results included measurements of their standard benchmark with the General Parallel File System (GPFS) for clusters, complementing earlier results with Lustre, AFS and other file systems. The AFS performance is rather poor for large files, but can be considerably improved by object extensions – the AFS client accesses the object store directly rather than through a single, bottleneck AFS server. CERN presentations were on Internet Small Computer System Interface (iSCSI) storage and Lustre evaluation.

Operating systems and applications

The presentations and discussions covered Scientific Linux (SL): the end of SL4 support (confirmed for October 2010), the status of SL5 and prospects for SL6. Addressing the need to support new hardware and provide more up-to-date releases of desktop software, FNAL is now offering FermiLinux Short Term Support, currently based on Fedora 10. Another talk described an alternative approach to pack Grid software for the most popular Linux distributions. Further presentations reported experience with Puppets (a configuration manager), Dovecot (an smtp mail server), OpenSharedroot (a tool to share a root file system across machines for high availability) and Slurm (a resource manager).

Security and networking

Talks described the CERN tools for detecting abnormal network behaviour, an SMS-based system to provide users with a one-time password required on top of the standard username/password combination for ssh access, and new requirements at FNAL when connecting to its network (an automatic software inventory is compulsory).

Data centres

This session saw two contrasting talks: one from a site that is launching a project to build a new data centre (they are full of optimism and faith) and one from a site that has just finished building a centre (their degree of optimism is quite a bit lower).

Conclusion

The meeting offered a broad spectrum of presentations, which led to intense discussions during the breaks. New opportunities for commonalities and collaborations were identified, and existing partnerships received a boost. The next meetings are scheduled for 26–30 October at LBNL in Berkeley, spring 2010 in Lisbon and autumn 2010 in the US.

Useful link

HEPiX website: www.hepix.org