Dryad: Distributed DataParallel Programs from Sequential Building Blocks

1
Dryad Distributed Data-Parallel Programs from
SequentialBuilding Blocks

• Michael Isard, Mihai Budiu, Yuan Yu,
• Andrew Birrell, Dennis Fetterly
• Microsoft Research, Silicon Valley

Eurosys 2007
2
Overview

• Mechanism to express parallel computation
• Large scale internet services
• Chip multiprocessors
• Drawing on previous work
• Condor
• GPU shader languages
• MapReduce
• Parallel databases

3
Main Idea

• Represent the computation as a DAG of
  communicating sequential processes

4
Claims

• The Dryad execution engine
• Schedules across resources
• Optimizes the level of concurrency in a node
• Manages failure resilience
• Delivers data where needed
• Performance is good and scales
• The programming abstraction is at the right level
• API mastery only takes a couple of weeks
• Higher level abstractions have been built on Dryad

5
Dryad System

• Name server enumerates all resources
• Including location relative to other resources
• Daemon running on each machine for vertex dispatch

6
Communication
7
Constructing the Job

• Use graph operators implemented in C to
  describe the graph.

8
Database Query Example
9
Execution

• Job manager not currently fault tolerant
• Vertices may be scheduled multiple times
• Each execution versioned
• Execution record kept- including versions of
  incoming vertices
• Outputs are uniquely named (versioned)
• Final outputs selected if job completes
• Non-file communication may cascade failures
• Vertices specify hard constraints or preferences
  for placement
• Scheduling is greedy assuming only one job

10
Run-time Graph Refinement
11
Results I

• SQL Query
• 10 Machines
• 2 dualcore 2 GHz
• 8 GB Mem
• 1 Gb Ethernet
• 4x400GB disks
• Winows Server 2003

12
Results II

• Map then reduce style
• Builds histogram of MSN Search query frequency
• 1800 Machines
• 10.2 TB source data
• 11072 Vertices

13
Refinement
14
Dryad