Key k has an old copy in a run and a new copy in the memory table. Which value wins?

• The oldest copy, it was written first
• A mix of both copies
• The newest copy in the memory table

Where does a fresh write go first?

• Into the oldest run on disk
• Directly into a compacted level
• Into the memory table

Compaction merges sorted runs and drops shadowed copies.

Circle one:   True   False

A workload writes constantly and rarely reads. Which policy fits?

• Tiered merging, which merges rarely
• Leveled merging, which merges often
• No compaction at all

A read looks for a key stored in three runs. What must it do?

• Check newest first and stop at the first copy found
• Check oldest first and return the last copy
• Read all three and average the values

A run membership filter says the key is absent. What should the read do?

• Scan the run anyway to be sure
• Skip the run entirely
• Compact the run first

Reads are slow and each key spans many runs. Which change helps most?

• Flush the memory table less often
• Move toward leveled merging plus filters
• Drop the membership filters to save memory

Pat says leveled merging rewrites data less than tiered merging. What is wrong?

• Leveled merging rewrites more, that is its price for short reads
• Tiered merging never merges at all
• Rewrites have no effect on write cost