Proceedings of the 29th International Symposium on High-Performance Parallel and Distributed Computing 2020
DOI: 10.1145/3369583.3392683
|View full text |Cite
|
Sign up to set email alerts
|

funcX: A Federated Function Serving Fabric for Science

Help me understand this report
View preprint versions

Search citation statements

Order By: Relevance

Paper Sections

Select...
3
1

Citation Types

1
55
0

Year Published

2020
2020
2023
2023

Publication Types

Select...
5
3
1

Relationship

1
8

Authors

Journals

citations
Cited by 153 publications
(56 citation statements)
references
References 21 publications
1
55
0
Order By: Relevance
“…The first phase of the pipeline is integrated with the Advanced Photon Source (APS) Data Management System at the beamline, which deposits each newly acquired image into a Globus-accessible storage system at the APS. As new images are acquired, Globus Automate flows are launched to process them as follows: 1) moves new files from APS to Theta by using the Globus Transfer service ( 35 ); 2) performs DIALS stills_process ( 36 ) on batches of 256 images by using funcX ( 37 ), a function-as-a-service computation system [funcX uses Parsl ( 38 ) to abstract and acquire nodes on Theta as needed, and dispatches tasks to available nodes]; 3) extracts metadata from files regarding identified diffractions and generates visualizations (funcX) showing the locations of positive hits on the mesh; and 4) publishes raw data, metadata, and visualizations to a portal on the Argonne Leadership Computing Facility (ALCF) Petrel data system ( 39 ). The result of this automated process is an indexed, searchable data collection that provides full traceability from data acquisition to processed data that can be used to inspect and update the running experiment.…”
Section: Methodsmentioning
confidence: 99%
“…The first phase of the pipeline is integrated with the Advanced Photon Source (APS) Data Management System at the beamline, which deposits each newly acquired image into a Globus-accessible storage system at the APS. As new images are acquired, Globus Automate flows are launched to process them as follows: 1) moves new files from APS to Theta by using the Globus Transfer service ( 35 ); 2) performs DIALS stills_process ( 36 ) on batches of 256 images by using funcX ( 37 ), a function-as-a-service computation system [funcX uses Parsl ( 38 ) to abstract and acquire nodes on Theta as needed, and dispatches tasks to available nodes]; 3) extracts metadata from files regarding identified diffractions and generates visualizations (funcX) showing the locations of positive hits on the mesh; and 4) publishes raw data, metadata, and visualizations to a portal on the Argonne Leadership Computing Facility (ALCF) Petrel data system ( 39 ). The result of this automated process is an indexed, searchable data collection that provides full traceability from data acquisition to processed data that can be used to inspect and update the running experiment.…”
Section: Methodsmentioning
confidence: 99%
“…These services are complementary, as they can be integrated into SCIFFS. Finally, we note works that demonstrate the benefits of serverless for data analytics (e.g., PyWren [29], IBM-PyWren [46], Flint [30], Locus [43], and funcX [19]).…”
Section: Related Workmentioning
confidence: 99%
“…While most previous work has focused on execution of serverless functions on homogeneous systems, Chard et al [4] propose funcX, a distributed function execution platform that supports various cloud platforms and modern HPC systems with underlying heterogeneous compute nodes. In contrast to our work, funcX does not support synchronous training of ML models, and is specifically designed for scientific computing.…”
Section: Related Workmentioning
confidence: 99%
“…These range from resource-constrained edge devices to modestly priced servers with mid-range resources to expensive highperformance computers with extensive compute, storage, and network capabilities. The introduction of serverless computing, particularly function-as-a-service (FaaS) [2] has made the execution of functions on these heterogeneous devices along with the cloud possible [4].…”
mentioning
confidence: 99%