• Posts
  • RSS
  • ◂◂RSS
  • Contact

  • Hit frequencies

    December 2nd, 2016
    tech  [html]
    When load testing server-side software there are two straight-forward options:
    1. Stress all endpoints evenly
    2. Stress one endpoint in isolation
    For example, if you're testing a dynamic website with siege you could (1) give it a list of all the urls on your site and have it rotate among them, or you could (2) pick a representive url and just hit that one repeatedly. Like any synthetic test this isn't completely realistic, but in this case you're missing something pretty important: if you want to test something that uses caching, you need to approximate the distribution of a real workload.

    Why? How well caching works depends on your hit rate, which depends on what the distribution of your requests looks like. Situation (1) is a caching worst-case, (2) is a caching best case, and your real situation is somewhere in the middle. For example, from my access logs over the past couple years, here's the distribution of requests I've seen:

    1275358 /news.rss
    562369 /
    437312 /favicon.ico
    406583 /index
    168296 /robots.txt
    162728 /p/
    116353 /simple_piano_recordings/piano_chords.svg
    94048 /icdiff
    92680 /p/mercury-spill
    63728 /wsgi/json-comments-cached/gp/i6WrmjSHXLg
    [snip ~1k entries]
    1179 /news/2011-11-02
    1178 /wsgi/json-comments/gp/YTnbXQoRPRN
    1178 /fiddle-clip-on/pictures/11-top.jpg
    [snip ~10k entries]
    113 /news/back_from_564.rss
    113 /news/2012-07-24
    113 /news/2012-07-09.html
    [snip ~100k entries]
    3 /nextbus/omnitrans/82/7387/
    3 /nextbus/art/1135/759763/
    3 /nextbus/mbta/39/6460/
    [snip ~1M entries]
    1 /ngx_pagespeed_beacon?ets=load:906&rload=1605...
    1 /nextbus/jtafla/17/1433/next/
    1 /news/all/trillion-dollar-platinum-coin

    This is kind of a mess, but the main idea is that we have a few "hot" endpoints, and then a long tail of less popular entries. I can use this distribution in load testing, to get something between uniform sampling (option 1) and singleton sampling (option 2) that better represents what real load on the site would look like.

    In case this is useful to other people, here's the frequency list (hits-frequency.txt.gz) and a short script to sample from it (generate_urls.py).

    Comment via: google plus, facebook

    Recent posts on blogs I like:

    Experiences in raising children in shared housing

    Sometimes I see posts about people’s hope to raise children in a group housing situation, and it often seems overly optimistic to me. In particular they seem to expect that there will be more shared childcare than I think should be expected. Today I talke…

    via The whole sky October 18, 2021

    What to learn

    It's common to see people advocate for learning skills that they have or using processes that they use. For example, Steve Yegge has a set of blog posts where he recommends reading compiler books and learning about compilers. His reasoning is basicall…

    via Posts on Dan Luu October 18, 2021

    EDT with updating double counts

    I recently got confused thinking about the following case: Calculator bet: I am offered the opportunity to bet on a mathematical statement X to which I initially assign 50% probability (perhaps X = 139926 is a quadratic residue modulo 314159). I have acce…

    via The sideways view October 12, 2021

    more     (via openring)


  • Posts
  • RSS
  • ◂◂RSS
  • Contact