Sparrow (CC, Distributed Comptuing)

Question 1

What is the Background (Sparrow)?

Answer 1

Sparrow is a distributed, low-latency scheduling system. It’s designed as an alternative to centralized scheduling.

Question 2

What is the Problem (Sparrow)?

Answer 2

• Traditional centralized schedulers struggle to provide millisecond-scale scheduling delay and support millions of task scheduling decisions per second.
• Modifying today’s centralized schedulers to support sub-second parallel tasks presents a difficult engineering challenge.

Question 3

What is the solution to the problem (Sparrow)?

Answer 3

• Sparrow employs a decentralized architecture with many parallel schedulers that don’t maintain state information.
• It utilizes batch sampling, extending it to prevent race conditions.
• Sparrow incorporates techniques inspired by Tarcil, a shared-state scheduler.

Question 4

What are the applications (Sparrow)?

Answer 4

The provided document does not explicitly detail real-life applications of Sparrow. However, it suggests applications in low-latency user-facing services like language translation or online search

Question 5

Strengths and Weaknesses (Sparrow)?

Answer 5

Advantages:
- Decentralization: Easy to add more schedulers.

• Load Balancing: Uses batch sampling.
• Fault Tolerance: Frameworks using Sparrow can automatically detect scheduler failures and failover to a backup scheduler in less than 120ms.
• Scalability: Designed for scaling to large clusters. Simulation results suggest it performs well as cluster size increases, even with tens of thousands of cores.
• Achieves median response times within 12% of ideal.
  It supports policies and constraints needed by analytics frameworks.

DisAdvantages: