Sync Computing - IT Press Tour #58 Oct. 2024

Transcript

1. AI for Data Infrastructure Sync Computing | 2024 Prepared for IT Press Tour Declarative Computing The future of automated cloud infrastructure
2. [Confidential] 01. 02. 03. 04. Agenda Introduction to Sync What is declarative computing? Introducing Gradient by Sync Looking ahead
3. [Confidential] Introduction to Sync Staff & Advisors From Founder’s story: Hailing from MIT / UC Berkeley with a background in high performance computing
4. Tech Partners • Databricks Technology Partner • NVIDIA Inception Program member • Native integrations for Airflow, Azure Data Factory, NVIDIA RAPIDS, etc.
5. [Confidential] 01. 02. 03. 04. Agenda Introduction to Sync What is declarative computing? Introducing Gradient by Sync Looking ahead
6. [Confidential] The Resource Allocation Problem: Today Cost: $100 Runtime: 1 hour Latency: 300 ms Compute Resources
7. Business Problems with Today’s Architecture Can’t miss SLA deadlines Compute cost are too high Cannot tune at scale
8. [Confidential] Declarative Compute Resources Cost: $50 Runtime: 1 hour Latency: 100 ms Cost: Minimum Runtime: 1 hour Latency: 100 ms
9. Reinforcement ML Model for Cloud Infra Key Concept: Closed-loop feedback enables automatic tuning towards infrastructure goals
10. [Confidential] 01. 02. 03. 04. 05. Agenda Introduction to Sync What is declarative computing? Introducing Gradient by Sync How are we different? Looking ahead
11. Introducing Gradient by Sync Can’t miss our SLA deadlines Teams are relying on mission critical pipelines finishing on time- hitting SLA goals is critical Cannot tune at scale Platform teams struggle to meet business demands due to lack the Spark expertise to make changes confidently Customer’s Problems Gradient’s Solutions Visibility to make teams faster Gradient’s insights and visibility helps data engineers quickly diagnose problems and get to solutions faster Free data engineering time With automatic cluster tuning, data teams are free to focus on more business relevant tasks, freeing up precious time Databricks cost are too high Databricks is a high percentage of cloud spend with ample opportunity to save costs Maximize cloud computing performance Gradient’s ML model custom trains and tunes each cluster to reduce costs and improve efficiency
12. How is Gradient Different? Cannot scale As you scale your data infrastructure, platform teams struggle to tune clusters to meet business demands due to expertise, or the sheer volume of workloads running (e.g. you cannot manually tune 10K jobs) Alternatives Gradient Opinionated optimization Leverage years of research and millions of DBUs managed that have shaped opinions about the right metrics to monitor and power Gradient’s intelligent insights & custom optimizations. Advanced ML models Gradient’s self-improving ML algorithms were developed at MIT. They use closed-loop feedback to continue to improve Passive recommendations Lists of optimizations that might have an impact, can only take you so far. Active management of data infrastructure Gradient automates compute optimization with ML-powered optimizations, customized per workload
13. Gradient’s Goals Lower costs Faster runtimes Hit your SLAs Save time with automation
14. User results cost savings 2:1 ROI The average starting ROI customers see from Gradient 2x Faster runtimes, while hitting mission critical SLAs
15. User Results in Production
16. Gradient Usage
17. [Confidential] 01. 02. 03. 04. Agenda Introduction to Sync What is declarative computing? Introducing Gradient by Sync Looking ahead
18. [Confidential] Databricks is Just The Start
19. Solutions to Today’s Architecture SLAs met at scale Optimized compute costs Automated tuning at scale
20. AI for your data infrastructure