guijarrocvs

Report on CVS Services:Central and LCG-dedicated services: 

Report on CVS Services: Central and LCG-dedicated services Sebastian.Lopienski@cern.ch Manuel.Guijarro@cern.ch CERN IT/PS/UI May 2004

Outline: 

Outline Overview Projects What do they offer? Architectures Advantages Drawbacks Implementation Figures Problems Future

CERN CVS Services  Overview: 

CERN CVS Services Overview hosting CERN-related software projects Central CVS service: created following a service request collection of user requirements architecture proposals implementation based on assigned resources in production since 08/2002, 71 projects and 12GB of data LCG dedicated CVS service: requested not to rely on AFS aligns, as much as possible, with Central CVS service setup in production since 02/2004, 22 projects and 1.5GB of data

CERN Central CVS Service  Projects: 

CERN Central CVS Service Projects

CERN CVS Services  What do they offer?: 

CERN CVS Services What do they offer? data integrity (mirror every hour, daily archiving) several access methods: Kerberos IV, SSH, pserver automatic CVS lock monitoring and reporting good performance web user interfaces (CVSWeb, ViewCVS, Bonsai); administration tools: Project creation, Access restrictions, service support (200 Remedy trouble tickets per year + Web documentation) service monitoring (UIMON, CVS access check, Archiving check,…)

ArchitecturesCentral Service vs LCG Service: 

Architectures Central Service vs LCG Service X.cvs.cern.ch

AdvantagesCentral Service vs LCG Service: 

Advantages Central Service vs LCG Service fastest possible access to the repositories (local file system) independent of AFS automatic and transparent load-balancing and fault tolerance (via an ISSCVS DNS alias that distributes CVS requests among a farm of four servers)

DrawbacksCentral Service vs LCG Service: 

Drawbacks Central Service vs LCG Service dependency on AFS and DNS constant mirroring may affect performance load-balance is on repository level (not request level) slave server down => no fault tolerance fail-over requires human intervention (for the time being)

ImplementationCentral Service vs LCG Service: 

Implementation Central Service vs LCG Service DNS relocation via ISSCVS ISSCVS: A central service queries some home made metrics in each CVS server Redirects every 30 secs Metrics: SSH  SNMP For more info: V. Bahyl AFS (NFS?) automatic AFS volume resizing (adapt to each repository needs). tools and web interfaces for project relocation (via DNS) CVS domain cvs.cern.ch data replication system (CVSUP) access to repositories and user home directories on a file-system level via NFS automatic relocation not wanted

FiguresCentral Service vs LCG Service: 

Figures Central Service vs LCG Service 71 projects 4 new projects per month Ave. Size= 170MB Top. Size > 4GB less than 12 hours/year downtime 22 projects Not growing Ave. Size= 70MB Top. Size > 270MB less than 12 hours/year downtime

# of projects trend: 

# of projects trend

Not so many problems: 

Not so many problems disk failures: some Western Digital 120GB disks models were faulty backplane installation (less vibration) computer centre: refurbishment (relocation of servers) power cuts several short network interruptions (mainly at non-working hours) partial interruptions: CERN firewall, AFS problems - affecting only some users

Future: 

Future provide Kerberos V access (CERN soon ready) slave server relocation for LCG hardware update (some nodes are 5 years old) automatic detection of inactive projects migrate monitoring from UIMON->Lemon ¾ of service manpower will disappear => no more dev.

Thank you!: 

Thank you! Any questions? More information at: http://cern.ch/cvs http://cern.ch/lcgcvs