Alert.png The wiki is deprecated and due to be decommissioned by the end of September 2022.
The content is being migrated to other supports, new updates will be ignored and lost.
If needed you can get in touch with EGI SDIS team using operations @ egi.eu.

Resource Centres OLA and Resource infrastructure Provider OLA reports

From EGIWiki
Revision as of 11:33, 12 December 2011 by Fergadis (talk | contribs) (→‎2011)
Jump to navigation Jump to search
Main EGI.eu operations services Support Documentation Tools Activities Performance Technology Catch-all Services Resource Allocation Security


Introduction

EGI Performance is measured using two parameters: Availability and Reliability (definition).

Availability/Reliability data is provided by the MyEGI portal. Note: GridView Availability/Reliability views are now obsolete.

Availability/Reliability are measured at a Resource Centre (RC) level and at a Resource infrastructure Provider (RP) level (for NGIs and EIROs).

SAM metric results are used for the calculation of Availability/Reliability:

  • Metrics for RC Availability/Reliability computation are those that are part of the WLCG_CREAM_LCGCE_CRITICAL profile
  • Metrics for RP Availability/Reliability computation (top-BDII) are those that are part of the ROC profile

Service Level Targets

For a Resource Centre(RC)

Is is mandatory that EGI certified Resource Centres provide a minimum monthly Availability/Reliability as specified below (see the RC Operational Level Agreement for details). Availability/Reliability statistics (OPS VO) are issued on a monthly basis.

minimum Availability 70%
minimum Reliabilty 75%
Condition for suspension Resource Centres which have an Availability of less than 70% for three consecutive months will be suspended, i.e. removed from the production infrastructure. Note. This new suspension policy was introduced in April 2011, to increase the original 50% threshold to 70%.
Condition for justification Resource Centres not providing minimum monthly performance (70% availability, 75% reliability) MUST provide justification through a GGUS ticket.

For a Resource infrastructure Provider (NGI/EIRO)

As of January 2012, it is mandatory that top-BDII services operated by NGIs provide a minimum availability of 99% (see the RP Operational Level Agreement for details). Availability/Reliability NGI reports are distributed monthly.

Note: Service Level Targets specified below will come into force as of Januwary 2011.

minimum top-BDII Availability 99%
minimum top-BDII Reliabilty 99%
Liability Resource infrastructure Providers not providing the minimum requested monthly performance for one month (99% Availability, 99% Reliability) MUST provide a service improvement plan.
  • See the list of NGIs' Top-BDIIs used for the Availability/Reliability computation.

Performance reports

2011

Jan|Feb|Mar|Apr|May|Jun|Jul|Aug|Sep|Oct|Nov re-computations are in progress...

2010

EGI-wide Availability and Reliability

It is available here (xls file, data from May 01 2010)

Underperforming/Suspended RCs

Process for quality verification

  • Generation of statistics

Availability and reliability statistics are automatically generated the first week of the month by the Availability Computation Engine (Gridview until May 2011) using the profile in pdf format and placed under [1]. An Excel version is available at [2]

  • Preliminary processing

Once the reports are generated, sanity checks are performed by EGI SA1 (Task TSA1.8). After this step is completed, statistics are uploaded into the EGI document server. Links to monthly statistics will be provided on a regular basis at this wiki page.

  • Publication

An announcement of the new results is distributed by EGI SA1 (TSA1.8) to the NGI Operations Managers mailing list. COD (TSA1.7) is responsible of supervising statistics by chasing NGIs to chase sites that need to provide comments in case thresholds are not met, and identifies sites eligible for suspension. This phase starts by filing a ticket to the COD Support Unit. The overall comments gathering process is handled through tickets.

  • Handling of sites below targets

For a site that misses availability/reliability targets but is not eligible for suspension:

  1. a child ticket is opened by the COD team and assigned to the respective NGI, asking for explanation to be given
  2. the explanation must be produced within 10 working days since the ticket is received by the site (please see known issues section [3]). Reminders and escalation is performed in accordance to COD escalation procedures [4].
  3. if the explanation is found satisfactory the ticket is closed
  4. conversely if the explanation is not given in due time, or the explanation is found inadequate, COD escalation procedure will be followed [5], with the site being suspended if neither site or NGI reply to the ticket
  5. the child ticket can then be closed
  6. the parent ticket will be closed when all child tickets have been closed.
  • Handling of sites that are eligible for suspension

For a site that is eligible for suspension:

  1. a child ticket is opened by the COD team assigned to appropriate NGI, notifying that the site will be suspended within 10 working days (please see known issues section [6])
  2. after the 10 days period passes during which normal COD escalation procedures apply [7], the site is suspended by COD unless the NGI has intervened or the EGI Chief Operations officer objects.
  3. in the case of NGI intervention, non suspension will occur if both the COD and COO agree on the reasoning provided by the NGI
  4. the child ticket closes either when the site is suspended or when suspension is canceled
  5. the parent ticket will be closed when all child tickets have been closed
  • Wiki follow up page

Sites that fail to provide explanations justifying the failure to meet OLA targets, or the explanation is found inadequate, as well as sites that are suspended, will be recorded in a wiki page [8]

  • Recomputation precedure

Should there be doubts about the validity of Availability/Reliability reports, a RC/NGI can request recomputations according to the procedure defined at [9]

Known issues and recommendations to NGIs

  1. ACE as Gridview in the past, is always calculating reliability and reliability of a site as soon as it shows up in GOCDB and in the BDIIs, regardless of its certification status. While processing the data in order to generate the availability/reliability report, ACE takes into account the Certification status of the site at that moment in order to decide if the site is certified and as a result it will show up in the report, or if it uncertified and it has to be excluded. Thus newly certified sites will get inaccurate Availability/Reliability figures for the month they were certified and all months before that. Because the Certification status history is not currently available in the operations tools, until a solution is implemented NGIs should check if they have sites affected by this issue and report it as explanation. More information at [10] and [11]. As of December 2010, Gridview had included a snapshot feature so availability takes into account the topology at the last day of the month. While it does not solve the problem completely, it reduces its impact. However ACE reports (used since May 2011) do not include the snapshot feature yet.
  2. The calculations performed by ACE always take into account the information system status and gocdb information at the time the calculation is performed, and not that of a certain checkpoint in the past. The implication of this is that any complete recalculation has the risk of altering the results for sites that had correct numbers in the first place. Thus until a solution is found, complete recalculations are avoided whenever possible, and errors are fixed on per site basis for those that have lower number than they should.
  3. Weighted availability is calculated by multiplying the number of logical CPUs a site published with the published HEPSPEC value. It is important that these numbers are correct, if HEPSPEC for a site is too high or too low (for example in case of mistake) the overall NGI wighted availability will be affected.

Operational Level Agreements

Resource Centre Operational Level Agreement

Resource infrastructure Provider Operational Level Agreement

Links

  • Definition of Availability and Reliability and related computation algorithm (paper)