Skip to content

Latest commit

 

History

History
46 lines (28 loc) · 1.33 KB

response_time_degradation.md

File metadata and controls

46 lines (28 loc) · 1.33 KB
Raw

Response Time Degradation

If we plot time as y and data size as x, then we can easily measure response time. If our plot has linear or polynomial line, then we may have response time degradation.

The culprits

The causes of response time degradation are:

  • Working set exceeding RAM.
  • Queries taking longer as the data set grows.
  • Growing pool of clients.
  • Unbounded array growth.
  • Excessive number of indexes.

Diagnose and solve the culprits

Working set exceeding RAM

Diagnose: Use mongstat to show data coming along.For example, execute following command to show time, dirty amount of memory, used amount of memory, number of insert, queue read and write, and active read and write: mongostat --port 27001 -o "time=T,dirty=dirty,used=used,insert=Inserts,qrw=qrw,arw=arw"

Solve:

  1. Increase cache size by set cacheSizeGB in configuration file to higher value.
  2. Or, adding more RAM.

Queries taking longer as the data set grows

Diagnose:

  1. Set profiling level to debug anything and set threshold: db.setProfilingLevel(2, <THRESHOLD>)
  2. Show the all slow operations: db.system.profile.find().pretty()
  3. Generate plot: mlogvis <LOG_PATH> --no-browser --out <HTML_FILE>
  4. Show the plot: open <HTML_FILE>

Solve:

  1. Create appropriate indexes to support queries.