Serverless Services for Scientific Cloud Computing

IberGrid 2022 Conference

​

Germán Moltó - gmolto@dsic.upv.es

Sebastián Risco - srisco@i3m.upv.es

Vicent Giménez - vigial@posgrado.upv.es

Miguel Caballer - micafer@i3m.upv.es

​

Agenda

• Motivation for Serverless Computing
• SCAR
• OSCAR
• MARLA
• TaSCaaS
• Summary

Motivation: Head Count

• ​

​

​

Motivation: Trusting Your Partners

• The Australian Bureau of Statistics, through open tender, awarded IBM a $9.6M a contract to implement an eCensus solution for 2016.
• ABS wisely tendered for services to “Perform Load Testing” ($469K out of which $325K was spent on software licenses).

​

​

Motivation: A Story in Three Acts

​

​

https://twitter.com/TurnbullMalcolm/status/762940763801989121

https://twitter.com/ABSStats/status/762961251764805633

​

https://twitter.com/narelleford/status/762984702915465216

Motivation: Official vs Unofficial

​

• Official Statement (13/10/2016) from the Office of Cyber Security Special Adviser:
• […] although the site withstood an initial DDoS attack and was coping with over 7,000 census forms a minute, a second and third attack took it down
• Critics: The system was believed to have been built on IBM WebSphere and run on IBM Softlayer (on-premises Cloud) instead of on a public Cloud.

Motivation: A Surprising Turn of Events

• A couple of students, without prior experience in AWS, developed a serverless system over a weekend supporting 4 times the workload used to test IBM’s system for $500 $30

​

https://eftm.com/2016/08/how-two-uni-students-built-a-better-census-site-in-just-54-hours-for-500-30752

Motivation: Standing on the Shoulders of Giants

• How could these be possible?
• Students had used AWS Lambda, a massively scalable serverless platform for event-driven computing.
• Serverless: Event-driven computation on a computing platform entirely managed by the Cloud provider

​

https://twitter.com/werner/status/765599106387542016

Motivation: Long Story Short

• IBM reportedly payed $30M to the Australian government as reports are released from two inquiries into DDoS attacks on census website.
• PwC Australia operated Australian 2021 Digital Census on (quick poll):

​

Serverless Computing

• Event-driven computing on highly-elastic services with fine-grained billing managed by the Cloud provider.

​

SCAR - https://github.com/grycap/scar

​

• Framework to execution Docker-based applications in AWS Lambda
• Highly-parallel event-driven file-processing serverless applications that execute on customized runtime environments provided by Docker containers run on AWS Lambda (thanks to uDocker)
• Pioneered the usage of Docker containers in AWS Lambda since 2017 (native support in AWS Lambda introduced end of 2020, now available in SCAR)
• Featured in the CNCF Cloud Native Interactive Landscape (>500 stars in GitHub): https://landscape.cncf.io/serverless?selected=scar
• Integrated with API Gateway for HTTP-based scalable endpoints
• Integrated with AWS Batch for cloud-bursting into scalable virtual HPC-based clusters (even with GPU support).

SCAR - Architecture

Pérez, A., Moltó, G., Caballer, M., & Calatrava, A. (2018). Serverless computing for container-based architectures. Future Generation Computer Systems, 83, 50–59. https://doi.org/10.1016/j.future.2018.01.022

SCAR - Sample Workflow for Multimedia Processing

• Lambda functions for audio processing
• AWS Batch jobs for video processing

Risco, S., & Moltó, G. (2021). GPU-Enabled Serverless Workflows for Efficient Multimedia Processing. Applied Sciences, 11(4), 1438. https://doi.org/10.3390/app11041438

OSCAR - https://oscar.grycap.net

• Open Source Serverless Computing for Data-Processing Applications (OSCAR)
• Serverless computing for Docker-based computationally-intensive applications on elastic Kubernetes clusters deployed on multi-Clouds.
• Mimics the event-driven computational paradigm of SCAR but for on-premises (or public) Clouds.

OSCAR - Architecture

• Dynamic provisioning of Kubernetes clusters on multiple Clouds thanks to the Infrastructure Manager (IM) - https://im.egi.eu
• Horizontally scalable Kubernetes clusters thanks to CLUES - https://github.com/grycap/clues

OSCAR - A Sample Use Case

• Mask Detection usage
• Combination of a cluster of Raspberry PIs (edge) and dynamically provisioned resources from EGI for AI inference along the computing continuum

https://oscar.grycap.net/blog/post-oscar-cloud-to-edge-approach-for-edge-ai-inference/

OSCAR + SCAR + EGI

• Functions Definition Language (FDL)

OSCAR - Interfaces

Deployment: Infrastructure Manager (IM) Dashboard: https://im.egi.eu

Web-based UI: https://ui.oscar.grycap.net

​

OSCAR CLI: github.com/grycap/oscar-cli

Composer: https://composer.oscar.grycap.net

MARLA - https://github.com/grycap/marla

• Deploy a serverless MapReduce processing engine on AWS Lambda.
• Files uploaded to Amazon S3 trigger the execution of the (parallel invocation of the) functions to concurrently process the dataset.

V. Giménez-Alventosa, G. Moltó, and M. Caballer, “A framework and a performance assessment for serverless MapReduce on AWS Lambda,” Futur. Gener. Comput. Syst., vol. 97, pp. 259–274, Aug. 2019, doi: 10.1016/j.future.2019.02.057

MARLA

• The experimental results unveiled that serverless platforms provide inhomogenous computing power that impacts coupled-computing executions of parallel jobs.

V. Giménez-Alventosa, G. Moltó, and M. Caballer, “A framework and a performance assessment for serverless MapReduce on AWS Lambda,” Futur. Gener. Comput. Syst., vol. 97, pp. 259–274, Aug. 2019, doi: 10.1016/j.future.2019.02.057

TaScaaS - https://github.com/grycap/tascaas

• Task Scheduler As A Service (TaScaaS) provides a complete serverless service to schedule and distribute High Throughput Computing (HTC) jobs among computational infrastructures.

V. Gimenez-Alventosa, G. Molto, and J. D. Segrelles, “TaScaaS: A Multi-Tenant Serverless Task Scheduler and Load Balancer as a Service,” IEEE Access, vol. 9, pp. 125215–125228, 2021, doi: 10.1109/ACCESS.2021.3109972.

Conclusions

• Event-driven computing allows to perform computing in response to events (such as file uploads) on a serverless platform which provides automated elasticity for dynamic resource provisioning.
• SCAR executes generic applications on AWS Lambda and automated extension to AWS Batch to allow elastic CPU/GPU batch computing on the Cloud.
• OSCAR implements the event-driven computing model of SCAR in on-premises Clouds, integrated with EGI services (EGI DataHub and EGI Federated Cloud).
• Innovative services can be developed on the foundations of serverless computing platforms, such as MARLA and TaSCaaS.

Contact

Germán Moltó

gmolto@dsic.upv.es

​

​

Instituto de Instrumentación para la Imagen Molecular (I3M)

Universitat Politècnica de València

@gmolto

Grant PID2020-113126RB-I00 funded by MCIN/AEI/10.13039/501100011033.

Project PDC2021-120844-I00 funded by MCIN/AEI/10.13039/501100011033 funded by the European Union NextGenerationEU/PRTR

Part of this work was supported by the project AI-SPRINT ‘‘AI in Secure Privacy-Preserving Computing Continuum’’ that has received funding from the European Union’s Horizon 2020 Research and Innovation Programme under Grant 101016577

​