Size distributions for memory functions under specific workloads

This folder contains a set of files that are included from

.

Offloading this data to individual files helps

• C++ editors (large arrays are usually not well handled by editors),
• and allows processing data by other tools to perform analysis or graph rendering.

Format

Most filenames are of the form

. They contain a single line of comma separated real values representing the probability that a particular size occurs. e.g.

• "0,1"
  indicates that only the size
  1
  occurs,
• "0.5,0.5"
  indicates sizes
  0
  and
  1
  occur with the same frequency.

These files usually contains sizes from

to

4096

inclusive. To save on space trailing zeros are discarded.

Workloads

As identified in the automemcpy paper:

• GoogleA
  <->
  service 4
• GoogleB
  <->
  database 1
• GoogleD
  <->
  storage
• GoogleL
  <->
  logging
• GoogleM
  <->
  service 2
• GoogleQ
  <->
  database 2
• GoogleS
  <->
  database 3
• GoogleU
  <->
  service 3
• GoogleW
  <->
  service 1

is an additional synthetic workload that simply returns a uniform repartition of the sizes from

384

to

4096

inclusive.

Note

Except for

, all distributions are gathered over one week worth of data.