Let No Good Deed Go Unpunished: Let No Good Deed Go Unpunished


The Optimal Good Deed:Centralized High Performance Research Computing Environment: The Optimal Good Deed: Centralized High Performance Research Computing Environment Researchers focus on their disciplines Efficiency (OS/patches/rebuild/security) NSF, etc requiring central support Leverage vendor relationships Consolidate licensing, expenses Consolidate space and environmental impact


Initiating the Good Deed: Initiating the Good Deed Provost was instrumental in selecting key faculty members to be interviewed Provost office established a Research Computing Steering committee Provost organized a central meeting open to all PIs and faculty members


Interview Process: Interview Process Varied random sample – but was cross discipline Questions pertaining to desired environments/platforms/configurations Clusters vs. SMP Immediate benefits and complexity Weighing Currently Uses vs. DREAMS Conflicting requests Vast software needs and desires


Interview Results: Interview Results Finding consistent terms Attempt to resolve conflicting concepts Translating terms into viable services Compatibility with IS strategies.


Hardware Desires : Hardware Desires Cluster With Job Management (queues) Ability to interactively run jobs Need to be able to support large files Need to be able to efficiently support large data sets (in excess of 4gig)


Software Desires: Software Desires C Matlab Mathmatica Fortran Java Maple MPI/Open MPI Numerical Algorithms Group EndNote Gaussian Portland Group Compiler C (Intel) Fortran (Intel) Stata SPSS Oracle SAS Amos AutoCad EQS TechPlot ChemDraw


Anticipated Cluster Configuration: Anticipated Cluster Configuration


To do an easy RFP: To do an easy RFP


RFP Responses: RFP Responses


Evaluation Process: Evaluation Process Hardware considerations: Intel vs. AMD Dual core? Dual Processor(s) Hyperthreading 64bit vs. 32bit Watts per thread SCSI vs. SATA OS/Configuration considerations: Software vendor support for the OS. Ease of install Uniqueness vs. Commodity


Evaluating RFP responses (cont): Evaluating RFP responses (cont) Onsite services Assembly Install Customization Administrative Training Administrative overhead Node rebuild Software distribution Software support: By 3rd party software vendors


Anticipating Support and User Acceptance: Anticipating Support and User Acceptance Commodity OS Support networks (user groups) Plenty of literature… Classes/training


The (easy) Decision … : The (easy) Decision … Hardware: 1 Front end – Dual processor – Xeon EMT64 4gig of Ram 240gb of disk space (3x80gig) 64 Compute nodes – Dual Processor – Xeon 4gig of Ram 40gig SATA hard drives Gigabyte networking between nodes


The Decision Continued…: The Decision Continued… OS: Platform Rocks Redhat Enterprise 3.0 WS User Environment: Platform Lava (Job Management) Sungrid Engine (job queues) Native Linux/Bash shell


IS buy-in: IS buy-in Earn extra buy in by having IS units answer the questions the RFP did not address: Account administration/Authentication User and Scratch Storage space Backup, Restore and DR strategies User Support/Issue workflow


Determining Datacenter Impact: Determining Datacenter Impact Heating Cooling Electrical Space Today Growth


Trigger for Other new services : Trigger for Other new services Datacenter hosting Advanced software/programming support Consultation on grant writing and equipment specifications


Support Impacts on existing services: Support Impacts on existing services HelpDesk Administrators Downtime response Software training and questions No additional FTE….


The Good Deed Decision: The Good Deed Decision Sort of ….


Maybe it’s a Good Deed: Maybe it’s a Good Deed Policy and Central IT View


Convincing IT : Convincing IT Leo had the easy part


Infrastructure : Infrastructure Politics in general Politics in specific just who is Academic Technology Services Who wants more work?


IS buy-in: IS buy-in Earn extra buy in by having IS units answer the questions the RFP did not address: Account administration/Authentication User and Scratch Storage space Backup, Restore and DR strategies User Support/Issue workflow


Support Redux: Support Redux HelpDesk Escalation Software installation Space


Governance: Governance Northeastern's Academic Plan Top 100 Watching the Competition Opportunity for a New Model


Setting Researchers Expectations: Setting Researchers Expectations Old Adage Don’t Promise What You Can’t Deliver


Scientists vs Those Other (social) Scientists: Scientists vs Those Other (social) Scientists The Good Deed Gets Murkier The coming issues Different software Processing vs. memory GUI vs. commands Lack of money Increased support requirements


Slide29: Discipline Research Stacks Scholarly Infrastructure


And the Most Punishing Part of the Deed: And the Most Punishing Part of the Deed Funding Ongoing funding


Acknowledgements: Acknowledgements To Associate Provost Srinivas Sridhar for his foresight in realizing the need to fund this cluster To IS VP Bob Weir for taking this bold step into new territory To the research faculty for realizing there is potential in this for all


Questions?: Questions? Leo Hill, Academic and Research Technology consultant, l.hill@neu.edu Leslie Hitch, Ed.D. Director Academic Technology, l.hitch@neu.edu Glenn Pierce, Ph.D., Senior Research Scientist, College of Criminal Justice, g.pierce@neu.edu