Download
evaluation of a new grid engine monitoring and reporting setup n.
Skip this Video
Loading SlideShow in 5 Seconds..
Evaluation of a new Grid Engine Monitoring and Reporting Setup PowerPoint Presentation
Download Presentation
Evaluation of a new Grid Engine Monitoring and Reporting Setup

Evaluation of a new Grid Engine Monitoring and Reporting Setup

149 Views Download Presentation
Download Presentation

Evaluation of a new Grid Engine Monitoring and Reporting Setup

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

  1. Evaluation of a newGrid EngineMonitoring and Reporting Setup Thomas Finnern

  2. Abstract for Conference • Title • Evaluation of a new Grid Engine Monitoring and Reporting Setup • Summary • Dashboards and Event Correlation for Grid Engine with Splunk • Content • Splunkis a commercial software platform for collecting, searching, monitoring and analyzing machine data providing interactive real-time dashboards integrating multiple charts, reports and tables. We have been working on a grid engine setup based on the free branch supporting standard reporting and simple job and fairshare debugging with easy chart generation and smart event correlation features. On top of this we try understand the added value of the enterprise branch supporting integrated user authentication and role-based access controls. There is a plan to share our work in a public available Splunk grid engine app.

  3. Outline • Grid Engine Data andSplunk • Integration intoSplunk • Job Data: submit, prolog, epilog plus Infos on Errors • SoGE (Son ofGrid Engine) Data: Messages andAccounting • System Data: Resources andUsage (numbers, projects) • Needs • Job Inspection (View 1) • Accountingand Reports (View 2) • Weekly Project Views (View 3) • Realtime System Data (View 4) • RoleBased Access • Grid Engine App … • Status andConclusions

  4. Integration intoSplunk • SplunkIndexing • Index: Separate Directory usedby all SoGEData • Index: Event indexing Time, Originating Host andSourcetype • ASCII Store for Data Sets • Field Mapping on theFly during Analysis • Splunk Data Input • Extra SyslogPort mappedSoGE Index • Submit, Jobstart, Jobend, System Data • SplunkForwarder • Running on Grid Engine Master • Reliable Upload toSplunk Server • Configured for Message File • Configured for Accounting File

  5. Grid Engine Data andSplunk

  6. Job Data via UDP Syslog • Setup • Perl Script in JSV(Job SubmitVerification on Server), Prolog and Epilog • Submit • eventtype=“sgelog” sgeevent=“submit" • sgeusersgejobidsgerootsgecellsgehost • Prolog, Epilog • eventtype=sgelogsgeevent={prolog|epilog} • sgeusersgejobidsgerootsgecellsgehostsgequeuesgeslotssgetaskidsgearchsgehostssgepesgesubmithost

  7. Job Inspection (View 1): Find activeusers, jobsandhosts • index=bird| transactionsgejobidstartswith=submit span=<time> | searchsgejobid=* sgeuser=* | chartvalues(sgeproject) values(sgeuser) values(sgejobid) values(sgehost) bysgeuser

  8. Job Data via TCP Splunk Forwarder • Setup • SplunkForwarder (same RPM as on Splunk Master) • Reliable Data CollectionoverSplunk Protocol for accountingandmessagesfiles • Grid Engine Messages • Running on qmaster, optionally on workernode • sourcetype=ge_messagessgeevent=„{I,W,E}“ sourcesgejobidsgettaskidsgemessagesgescope • Grid Engine Accounting • Running on qmaster • sourcetype=ge_accountingsourcesgejobidsgedistrosgeusersgear_submission_timesgearidsgecpusgedepartmentsgeend_timesgeexit_statussgefailedsgegranted_pesgegroupsgehostsgeiosgeiowsgejobnamesgemaxvmemsgememsgepe_taskidsgeprioritysgeprojectsgequeuesgeru_idrsssgeru_inblocksgeru_ismrsssgeru_isrsssgeru_ixrsssgeru_majfltsgeru_maxrsssgeru_minfltsgeru_msgrcvsgeru_msgsndsgeru_nivcswsgeru_nsignalssgeru_nswapsgeru_nvcswsgeru_oublocksgeru_stimesgeru_utimesgeru_wallclocksgeslotssgestart_timesgesubmission_timesgetask_number

  9. Reports andAccounting (View 2): GridEngine Accounting(30d) • Pie „Sum CPU Secondsby Project“ • Timechart „ Sum CPU Secondsby Project“ • Queue Timing „Wait Times and Wall Times by Queue“ • Timechart „Jobs in Error by Project“ • Table „All Values“

  10. Accounting Analysis

  11. System Data via UDP Syslog • Setup • PerlscriptrunningCommandsqhostandqstat in Cron Job on Master and Slave • qhostprovides Worker Node Resources, qstatshows Project Data • sgeevent=„numbers“ • sgelog: sgehost="global" sgeShareTotalsgeSumProject-<ProjectName> sgeSumJobs-Error sgeSumJobs-Waiting sgeSumJobs-SlotRunsgeSumJobs-RunningsgeSumShares-<ProjectName> sgeSumShare-Store sgeSumShare-Memory sgeSumShare-Cores sgeSumShare-Hosts sgeSumCores-Total sgeSumCores-Available sgeSumDistro-<Name> sgeSumQueue-l<QName> sgeSumQueue-total sgeSumStore-mem_usedsgeSumStore-h_vmemsgeSumStore-h_ftotalsgeSumStore-mem_totalsgeSumStore-h_fused • sgeevent=„projects“ • sgelog: sgehost="global" sgeprojectsgesharesgestcktsgeovrtsSlotInErrorsgeotcktsgetcktssgeftcktJobsInErrorSlotRunningsgememJobsRunningsgeiosgecpuJobsWaiting

  12. Weekly Project View (View 3):Project Infos plus System Infos • Pies „Slots, CPU, IO and MEM by Project“ • Timechart „Slots, Waiting Jobs and Tickets by Project“

  13. Realtime System Data (View 4): System Healthand Trends • Last 15 Minutes Jobs, Slots, Waitsand Errors • Timechart 24 HoursJobs, Slots, Waitsand Errors • Timechart „Core Usageby Queue“ • Sparkling Lines: Trends for Unavailablecores, Distros, Memory Usage • Table „All Summed Values“

  14. Enterprise vs. Free Version • Price per Data Volume… • Limit / License Handling • More Data (> 500 M) • RoleBased Data Access • Auto Report • HA • Limit on Data Volume … • Limit / License Handling • 500 Mbyte • Interactive Analysis • Overall Correlations • Easy Debugging • Puppet Install

  15. Status and Conclusion • Status • Work in Progress • Finetuning Reports • Checking Data Consistency • Still Learning Splunk • … • Conclusions • Wouldliketobuy • RoleBased Data Access • High Availibilty • … • Thankyou for Listening