Motivation for `Paper`

[shubham-sudo]

Comments (25th-Oct MOM)

• Assume the whole metadata is in-memory
• Run the experiment for LSM on the emulator
• The x-axis is the % of updates and the y-axis is the amplification
  • start with 0 % updates and go slowly till 100 % and then go up to 1000 %

• We just need to count the number of times the data page is accessed

• The write amplification in the B+ tree is for every insert

• The read amplification in the B+ tree is the number of data pages read during the whole workload

• The space amplification in the B+ tree is more due to the 50% full factor in each node.

• The write amplification in the LSM tree is on every flush plus due to the compactions happening in the background

• The read amplification in LSM is the number of more pages read because of updates. (We might read some of the logically invalid data)

• The space amplification is also because of invalid keys stored in the DB.

Result for Motivation

• One graph with read and write amplification for LSM, B+, and B-epsilon
• One graph for the space amplification for LSM, B+, and B-epsilon

[shubham-sudo]