![Page 1: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/1.jpg)
Introduction to DHTCBrian Lin
OSG Software TeamUniversity of Wisconsin - Madison
![Page 2: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/2.jpg)
OSG User School 2018
Local High Throughput Computing
localresources
compute
UW - Madison
2
![Page 3: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/3.jpg)
OSG User School 2018
Local High Throughput Computing
localresources
compute
UW - Madison
3
![Page 4: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/4.jpg)
OSG User School 2018
How do you get more computing
resources?
4
![Page 5: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/5.jpg)
OSG User School 2018
#1: Buy Hardware
5
![Page 6: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/6.jpg)
OSG User School 2018
#1: Buy Hardware
• Great for specific hardware/privacy requirements• Costs $$$
− Initial cost− Maintenance − Management − Power and cooling
• Delivery and installation takes time• Rack/floor space• Obsolescence• Plan for peak loads, pay for all loads
6
![Page 7: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/7.jpg)
OSG User School 2018
#2: Use the Cloud
7
![Page 8: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/8.jpg)
OSG User School 2018
• e.g. Amazon Web Services, Google Compute Engine, Microsoft Azure, Rackspace
• Fast spin-up• Costs $$$• Still needs expertise + management
− Easier than in the past with the condor_annex tool• Does payment fit with your institutional or grant
policies?
#2: Use the Cloud - Pay per cycle
8
![Page 9: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/9.jpg)
OSG User School 2018
• e.g. Cycle Computing, Globus Genomics• Pay someone to manage your cloud resources —
still costs $$$• Researchers and industry have used this to great
success− Using Docker, HTCondor, and AWS for EDA Model Development− Optimizations in running large-scale Genomics workloads in
Globus Genomics using HTCondor− HTCondor in the enterprise− HTCondor at Cycle Computing: Better Answers. Faster.
#2: Use the Cloud - ‘Managed’ clouds
9
![Page 10: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/10.jpg)
OSG User School 2018
#3: Share Resources
10
![Page 11: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/11.jpg)
OSG User School 2018
#3: Share Resources - Distributed HTC
UW - Madison
11
University of Nebraska - Lincoln
University of Chicago
![Page 12: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/12.jpg)
OSG User School 2018
i.Split Up Your Jobs Manually
Let’s start sharing!
12
![Page 13: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/13.jpg)
OSG User School 2018
Manual Job Split
• Obtain login access• Query each cluster for
idle resources• Split and submit jobs
based on resource availability
13Photo by Denys Nevozhai on Unsplash
![Page 14: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/14.jpg)
OSG User School 2018
#3: Share Resources - Distributed HTC
UW - Madison
14
University of Nebraska - Lincoln
University of Chicago
![Page 15: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/15.jpg)
OSG User School 2018
#3: Share Resources - Distributed HTC
UW - Madison
15
University of Nebraska - Lincoln
University of Chicago
![Page 16: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/16.jpg)
OSG User School 2018
Manual Job Split - Shortcomings
• Fewer logins = fewer potential resources• More logins = more account management• Why would they give you accounts? Are your friends
going to want CHTC accounts?• Querying and splitting jobs is tedious and inaccurate• Not all clusters use HTCondor — other job
schedulers e.g., SLURM, PBS, etc.• Pools are independent — workflows must be
confined to a single pool16
![Page 17: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/17.jpg)
OSG User School 2018
ii.Split Up Your Jobs Automatically
Let the computers do the work
17
![Page 18: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/18.jpg)
OSG User School 2018
Automatic Job Split - Shortcomings
Homer: Kids: there's three ways to do things; the right way, the wrong way and the Max Power way!Bart: Isn't that the wrong way?Homer: Yeah, but faster!
Groening, M (Writer), Michels, P. (Director) . (1999). Homer to the Max [Television Series Episode]. In
Scully, M. (Executive Producer), The Simpsons. Los Angeles, CA: Gracie Films 18
![Page 19: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/19.jpg)
OSG User School 2018
Automatic Partitions - Shortcomings
19Source: https://xkcd.com/1319/
![Page 20: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/20.jpg)
OSG User School 2018
#3: Share Resources - Requirements
• Minimal account management• No job splitting• DAG workflow functionality• HTCondor only!• No resource sharing requirements
20
![Page 21: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/21.jpg)
OSG User School 2018
iii.Overlay Systems
Let the OSG do the heavy lifting
21
![Page 22: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/22.jpg)
OSG User School 2018
The OSG Model
22
OSG
OSG Submit and CM
Cluster
![Page 23: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/23.jpg)
OSG User School 2018
The OSG Model
23
OSG Submit and CM
OSGCluster
![Page 24: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/24.jpg)
OSG User School 2018
The OSG Model
24
OSG Submit and CM
OSG
Pilot Jobs
Cluster
![Page 25: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/25.jpg)
OSG User School 2018
The OSG Model
25
OSG Submit and CM
OSG
Pilot Jobs
Cluster
![Page 26: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/26.jpg)
OSG Summer School 2017
Job Matching
• On a regular basis, the central manager reviews Job and Machine attributes and matches jobs to slots.
submit execute
execute
execute
central manager
![Page 27: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/27.jpg)
OSG User School 2018
The OSG Model
27
OSG Submit and CM
OSGCluster
![Page 28: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/28.jpg)
OSG User School 2018 28
The OSG Model - Jobs in Jobs
Photo Credit: Shereen M, Untitled, Flickr https://www.flickr.com/photos/shereen84/2511071028/ (CC BY-NC-ND 2.0)
![Page 29: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/29.jpg)
OSG User School 2018
#3: Share Resources - Requirements
• Minimal account management: only one submit server
• No job splitting: only one HTCondor pool• DAG workflow functionality: Only one HTCondor
pool• HTCondor only: Only one HTCondor pool• No resource sharing requirements: the OSG doesn’t
require that users “pay into” the OSG
29
![Page 30: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/30.jpg)
OSG User School 2018
The OSG Model - Recap
• Pilot jobs (or pilots) are special jobs• Pilots are sent to sites with idle resources• Pilot payload = HTCondor execute node software• Pilot execute node reports to your OSG pool• Pilots lease resources:
− Lease expires after a set amount of time or lack of demand− Leases can be revoked!
30
![Page 31: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/31.jpg)
OSG User School 2018
The OSG Model - Leasing the Cloud
• What if there aren’t enough idle resources?• Combine overlay system with cloud technology• Some of your OSG jobs may run in the cloud in the
next few years• … but this should be completely transparent to you
31
![Page 32: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/32.jpg)
OSG User School 2018
The OSG Model - Collection of Pools
• Your OSG pool is just one of many
• Separate pools for each Virtual Organization (VO)
• Your jobs will run on the OSG VO pool
32Photo by Martin Sanchez on Unsplash
![Page 33: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/33.jpg)
OSG User School 2018
The OSG Model - Getting Access
• During the school:− OSG submit node at UW (exercises)− OSG submit node via OSG Connect (Thursday)
• After the school:− Both of the above− VO-hosted submit nodes− Institution integration with the OSG
33
![Page 34: Introduction to DHTC · OSG User School 2018 • e.g. Cycle Computing, Globus Genomics • Pay someone to manage your cloud resources — still costs $$$ • Researchers and industry](https://reader036.vdocuments.site/reader036/viewer/2022070916/5fb6cfae8f7f890677749b98/html5/thumbnails/34.jpg)
OSG User School 2018
Questions?
34