Twine (CC, Resource Management)

Question 1

Background (Twine)?

Answer 1

• Twine is a cluster manager developed by Meta, deployed in 2011, and has expanded to 15 data center locations
• It’s designed for high availability and reliability.

Question 2

Problem (Twine)?

Answer 2

• Traditional cluster management systems often rely on static machine partitioning, which can lead to inefficient resource utilization.
• Customizing shared infrastructure and efficiently allocating resources are significant challenges in large-scale cluster management.
• Balancing performance, reliability, and cost-efficiency in a massive, geographically distributed infrastructure is complex.

Question 3

Solution (Twine)?
(TCMR)

Answer 3

• Twine uses dynamic machine partitioning instead of static clusters.
• Twine adopts smaller machines over larger ones for better power efficiency.
• It uses a Resource Allocation System (RAS) for efficient machine allocation.

Question 4

Applications/Uses (Twine)?

Answer 4

Twine orchestrates containers to run nearly all of Meta’s services.

Question 5

Strengths and Weaknesses (Twine)?

Answer 5

Strengths:
- Scalability: Twine is designed to manage millions of servers across multiple data centers.

• Availability and Reliability: It uses sharding and replication for high availability and reliability.
• Efficiency: Dynamic machine partitioning and shared infrastructure contribute to efficient resource utilization.
• Customization: Twine allows for host-level customization through entitlements.
• Collaboration: The TaskControl API enables applications to collaborate with the cluster manager.
• Twine is designed to be power-efficient.

Weaknesses: