Lecture 6 Flashcards
1
Q
What’s the original MapReduce mitigation approach for Stragglers?
A
- Run a speculative copy (called backup task)
- The copy or original would finish first
- Without speculative execution, the job would be slower but would improve the job response times
2
Q
When a node has an empty slot, Hadoop chooses one from the 3 categories in priority:
A
- Failed tasks are given higher priority
- Unscheduled tasks with local data to the node are chosen first
- Run speculative task
3
Q
Describe the progress score used to detect stragglers
A
- Progress score begins from 0…1
- *For mappers:** the score is the fraction of input data read
- *For reducers:** the execution is divided into 3 equal phases:
- Copy phase: percent of maps that output has been copied from
- Sort phase: map outputs are sorted by key: percent of data merged
- Reduce phase: percent of data passed through the reduce function
4
Q
Describe the Scheduler’s Assumptions
A
- Nodes can perform work at the same time/concurrently
- Tasks progress at a constant rate throughout time
- No cost starting a speculative task
- Too many speculative tasks can take away resources from other running tasks
- Tasks finish in “waves” so tasks with low progress score is likely a straggler
- Task’s progress is equal to the fraction of the total work
5
Q
Describe the LATE scheduler
A
Speculatively execute the task with largest estimated finish
“Longest Approx. Time to End”
Look forward than backwards
6
Q
Sanity thresholds of LATE scheduler
A
- Cap number of backup tasks
- Launch backups on fast nodes
- Only back up tasks that are sufficiently slow
7
Q
What is Hadoop?
A
- An open-source framework for intensive distributed applications
- Inspired by Google’s MapReduce + GFS
- Implemented in Java
8
Q
Describe the Hadoop master and slaves
A
- One machine in the cluster is designated as ‘NameNode’ or ‘JobTracker’.
- The rest of the other machines are in the cluster are ‘DataNode’ or ‘TaskTracker’ – “slaves/worker nodes.”