Presented By
O’Reilly + Cloudera
Make Data Work
29 April–2 May 2019
London, UK

A Magic 8 Ball for optimal cost and resource allocation for the big data stack

Shivnath Babu (Unravel Data Systems | Duke University), Alkis Simitsis (Micro Focus)
16:3517:15 Wednesday, 1 May 2019
Data Science, Machine Learning & AI
Location: Capital Suite 15/16
Average rating: *****
(5.00, 1 rating)

Who is this presentation for?

  • Data engineers, architects, and data scientists

Level

Intermediate

Prerequisite knowledge

  • Familiarity with Spark, Kafka, Flink, HBase, and Flink

What you'll learn

  • Learn how to build a decomposable time series model for optimal cost and resource allocation for the big data stack

Description

Many applications in fields like healthcare, genomics, financial services, self-driving technology, telecommunications, ads services, government, and media are being built on what is popularly known today as the big data stack. What’s unique about the big data stack is that it’s composed of multiple distributed systems and almost every application interacts with multiple distributed systems. For example, an SQL query may interact with Spark for its computational aspects, with YARN for its resource allocation and scheduling aspects and with HDFS or S3 for its data access and I/O aspects. Or a streaming application may interact with Kafka, Flink, and HBase. The nature of such distributed applications is that they interact with many different components that could be independent or interdependent—what’s often referred to in popular literature as “having many moving parts.” Nonetheless, besides the extremely high complexity of the big data stack systems, enterprises need to be able to provision for resources, usage, cost, job scheduling, and so on.

Prediction modeling lets you see the future. Making accurate predictions depends on having the right monitoring data and the right model. Enabling all the monitoring data in the big data stack to be collected and stored in a single place opens up interesting opportunities to apply statistical analysis and learning algorithms to this data. These algorithms can generate insights that, in turn, can be applied manually by the user or automatically.

Shivnath Babu and Alkis Simitsis detail how to build a Magic 8 Ball for the big data stack—a decomposable time series model for optimal cost and resource allocation that offers enterprises a glimpse into their future needs and enables effective and cost-efficient project and operational planning.

Topics include:

  • How an application performance management platform can collect the data needed to enable accurate predictions
  • How to use a decomposable time series model that utilizes a Bayesian-based curve-fitting method, which supports seasonality, to build interactive prediction features such as disk, CPU, and memory capacity forecasting and workload forecasting per user, tenant, project, queue/pool, and cluster
  • Cost and performance models to translate app and cluster level metrics to dollar cost and hardware resources
  • How such modeling can be used as a what-if analysis tool for a variety of use cases including cost charge-back, budget planning, resource provisioning, team and project planning, and cluster and cloud migration
Photo of Shivnath Babu

Shivnath Babu

Unravel Data Systems | Duke University

Shivnath Babu is the CTO at Unravel Data Systems and an adjunct professor of computer science at Duke University. His research focuses on ease of use and manageability of data-intensive systems, automated problem diagnosis, and cluster sizing for applications running on cloud platforms. Shivnath cofounded Unravel to solve the application management challenges that companies face when they adopt systems like Hadoop and Spark. Unravel originated from the Starfish platform built at Duke, which has been downloaded by over 100 companies. Shivnath has won a US National Science Foundation CAREER Award, three IBM Faculty Awards, and an HP Labs Innovation Research Award.

Photo of Alkis Simitsis

Alkis Simitsis

Micro Focus

Alkis Simitsis is a chief scientist for cybersecurity analytics at Micro Focus. Alkis has more than 15 years of experience building innovative information and data management solutions in areas like real-time business intelligence, security, massively parallel processing, systems optimization, data warehousing, graph processing, and web services. He holds 26 US patents and has filed over 50 patent applications in the US and worldwide. He’s published more than 100 papers in refereed international journals and conferences (top publications cited 5,000+ times) and frequently serves in various roles in program committees of top-tier international scientific conferences. He’s also an IEEE senior member and a member of the ACM.