ATLAS Edinburgh ECDF Project Tasks

Middleware Service List

Service Deployment

Virtualisation Requirements

Server estate

• 8 hypervisors (with 1 failover) will be enough to host all middleware services required over the next six months
• Replacement servers are provided upon host failure (with a pre-defined replacement window).

Server/Service Access

• Administrator access and console control of each hypervisor.
• Administrator access on all virtual instances.
• Identify any "opt-out" areas where admin access cannot given outside of systems team.
• Ability to power cycle hypervisors and all virtual instances outside of OS.

Hypervisor features

• Power management of running instances.
• Ability to create new instances.
• Ability to snapshot existing instances.
• Restore image from backup/snapshots.
• Live migration (optional).
• Live tuning of memory and CPU resources (optional).
• Export monitoring data to nagios and ganglia.
• Network accessible from grid-admin and eddie.
• Has access to a snapshot/backup repository.

Instance features

• Access to GPFS via NFS mounts (work (rw), applications (ro)).
• All network accessible from grid-admin and eddie.
• Network accessible from each of the virtual instances
• Keep storage of instance lightweight, but enough space for logs.
• Log backup/rotation to dedicated backup area (optional).
• Instances accessible from ppe network (optional).

Backup and Restore

• Ability to make on-demand backup for each service.
• Ability to make scheduled backup for each service.
• Retain backup files in central repository (up to a certain date/space limitation).
• Restore will be run only on host failure / outage > 12 hours.
• Rapid restore of core services on failover hypervisor
• No issues with host/IP conflicts or firewall upon instance restore

Future tasks

• Deploy and test ECDF VM solution
• Dedicated NFS (or CVMFS) host/service
• Consolidate site monitoring
• Configuration management service
• Change management (practice for reverse all changes)
• Intervention logbook
• Documentation!
• Multi-user pilot jobs
• ArcCE testing and production
• ATLAS analysis readiness
• Job efficiency studies
• Production squid service

ECDF/GridPP research projects

• CVMFS performance testing
• Advertise GPU queue on the grid
• Whole node Worker Nodes for multicore studies.
• High memory server testing

Processes

Open questions

• Are other virt services (mwvm01, mwvm02, Galaxy) still being run on mw04?
• Are admin03, mw01, mw02, mw03, grid2 available for gridpp use in the near future?

-- AndrewWashbrook - 17-Nov-2010