User Support, Campus Integration, OSG XSEDE Rob Gardner OSG - - PowerPoint PPT Presentation

User Support, Campus Integration, OSG XSEDE

Rob Gardner OSG Council Meeting June 25, 2015

• 149. Present to Council 1 page document on

"Enabling Campus Resource Sharing and use of remote OSG resources in 15 minutes - Rob Gardner, Frank

Enabling Campus Sharing & Use of OSG

• Clemson helping drive this development
• Two track strategy to integrate Palmetto

resource and user community

○ Track 1: “light, quick” ■ Sumit from Palmetto to OSG, and back ■ “Quick Connect” → OSG Connect to Palmetto via hosted Bosco service (ssh) ○ Track 2: “full OSG capabilities” ■ Full HT Condor CE, OASIS+Squid, {StashCache}

Working document: goo.gl/9aNkJs

Track 1: OSG Connect to Clemson-Palmetto

• Hosted service @ OSG Connect
• Addressed OrangeFS + Condor file locking

Current opp limit on Palmetto Capped at 500 jobs due to PBS Pro limitation that prevents Clemson users in the general pool from (non-owners) preempting OSG users. Expect a fix in next release of PBS Pro so that OSG jobs can claim additional idle cycles

• n Palmetto

submitted from login.osgconnect.net

Track 1: Submit from Clemson-Palmetto

• Download Campus Connect client from github
• Minutes to submission to OSG

Campus Connect Client

• lightweight module to

manage submission from a campus login host

• heavy lifting done at

hosted schedd

• In Year 4, extend to reach,

monitor, and account: ○ local campus allocation ○ XD allocation ○ Full integration with campus IDM & signup

Evaluating at:

Longer term: Hosted Campus CE-ssh

• Discussions to establish approach for hosted

CE services on behalf of campuses short of manpower

• Quick(er) on-ramp of a campus HPC cluster

without requiring local OSG expertise

• @ the campus: provide ssh access, local

accounts for supported VOs

• Normal CE operations handled by OSG staff
• Possible “umbrella CE” for small campuses

• 152. Pay attention to "Sound Bites" that

communicate the scale and reach of OSG to outside agencies/projects - Rob G, Bo, Clemmie

Open Science Grid: HTC supercomputer

• 2014 stats

○ 67% size of XD, 35% BlueWaters ○ 2.5 Million CPU hours/day ○ 800M hours/year ○ 125M/y provided opportunistic

• >1 petabyte data xfer/day
• 50+ research groups
• thousands of users
• XD service provider for XSEDE

Rudi Eigenmann Program Director Division of Advanced Cyberinfrastructure (ACI) NSF CISE CASC Meeting, April 1, 2015

Lowering barriers to usability

OSG as a campus research computing cluster

★ Login host ★ Job scheduler ★ Software (modules) ★ Storage ★ Tools

Software & tools on the OSG

• Distributed software file system
• Special module command

○ identical software on all clusters

• Common tools & libs
• Curate on demand

continuously

• HTC apps in XSEDE

campus bridging yum repo $ switchmodules oasis $ module avail $ module load R $ module load namd $ module purge $ switchmodules local

http://goo.gl/TlLq1M

Modules now used at most sites

All software accesses by module are monitored centrally for support purposes

User Tools

• $ tutorial
• $ connect

○ on login.osgconnect.net, on campus, or laptop

• $ module (software, all OASIS enabled sites)
• $ stash-cp (Stash to job, in development)

Education and Training assets

• Helpdesk with community forum and

knowledge base

• github seen as strategy for formal

management of user documentation

○ Markdown tutorials → same place as code ○ tutorial write-ups track code samples closely ○ auto html and upload to help desk (in seconds)

• Expect to announce helpdesk support.
• pensciencegrid.org this week

Code and Markdown managed in Github Content indexed, searchable

User sees personal history of support requests to OSG .. and can drill down to see full interaction

• history. Staff can

make private notes,

• r link to a Jira issue

for technical support tracking. Of course, all available via email: user- support@openscienc egrid.org Can DM tweet to @osgusers which generates a ticket

Uber-like feedback is collected (except we don’t rate users :)

• Software Carpentry includes a section on scientific

programming using Python. IPython Notebook is used for instruction.

• SWC typically asks users to install IPython on laptops; this

is a top source of delays and confusion.

• In our DHTC edition of SWC, we already have a multiuser

server with login accounts that users retain indefinitely.

• Idea: use this framework to provide a shared IPython,

establishing a common baseline for the toolchain.

IPython Notebooks

IPython Notebook Service

Developed a platform to launch per-user IPython Notebook servers:

➥

1. User visits http://ipython.osgconnect.net and logs in.

IPython Notebook Service

• 2. Server launches pre-configured

IPython Docker container. Docker provides user and data isolation. Containers can be shut down and re-instantiated on demand.

• 3. Within moments, a newly

provisioned IPython instance is

• available. Notebook storage is

persistent and accessible via login.osgconnect.net.

Education and Training activities

• Working with Tim Cartwright and Lauren Michael (ACI-REF)

to support 2015 OSG User School

• UChicago-Northwestern roundtable (postponed to “Fall”)
• OSG-SWC @ Duke, October 26-29 (tentative dates)

Joint Software Carpentry & Open Science Grid Workshop at Duke University Distributed high throughput computation is concerned with using many computing resources potentially spread over large geographic distances and shared between organizations. These could be university research computing clusters, national leadership class HPC facilities, or public cloud resources. Incorporating these into science workflows can dramatically benefit your research program. However, to get the most of these systems requires some knowledge and skill in scientific computation. This workshop extends basic instruction on Linux programming from the Software Carpentry series with concepts and exercises on distributed high throughput computation. Participants will use resources of the Scalable Computing Support Center as well as the Open Science Grid, a national supercomputing-scale high throughput computing facility. There will be experts on hand to answer questions about distributed high throughput computing and whether it is a good fit for your science.

Other Campus Outreach Events

• Internet2 Technology Exchange, October 4-7,

Cleveland (formal decision next week)

○ Distributed High Throughput Computation: a Campus Roundtable Discussion (Research Track)

• Rocky Mountain Advanced Computing

Consortium, HPC Symposium (Aug 11-13, Boulder)

○ 30 minute slot shared with XSEDE

• XSEDE15, CLUSTER15 (Campus Bridging)

OSG as XD Provider to XSEDE

OSG XD - Last 12 months

OSG XD: June XRAC Meeting (Nashville)

• OSG pledges 2M CPU-hours (SUs) per quarter
• There were 199 requests for XSEDE resources, mostly for

Stampede and Comet

• There were no requests for OSG resources
• Post meeting, following granted:

○ 50k SU to Kettimuthu/ANL (CS: workflow modeling) ○ 100k SU to Qin/Spellman (class on gene networks) ○ 1.39M SU to Gull/UMich (PHYS: condensed matter)

• Many NAMD requests

○ → start a MD-HTC activity with ACI-REF?

Conclusions & Outlook

• “Clemson on the air”

○ Local submit to OSG validated at scale of 1500 jobs ○ Joint use of campus and OSG resources in same work environment ■ Model for other campuses, ACI-REF as channel ○ Quick Connect to share resources is functional

• HTC training materials now formally managed
• Helping users via XSEDE

○ plan detailed studies of common application scaling properties & potential conversion to HTC workflow

user-support@opensciencegrid.org @osgusers