Skip to content

Commit

Permalink
docs: update throttling (#8854)
Browse files Browse the repository at this point in the history
  • Loading branch information
paulirish committed May 28, 2019
1 parent a47ab62 commit 10d9c6f
Showing 1 changed file with 10 additions and 10 deletions.
20 changes: 10 additions & 10 deletions docs/throttling.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,22 +7,22 @@ This is the standard recommendation for mobile throttling:
- Throughput: 1.6Mbps down / 750 Kbps up.
- Packet loss: none.

These exact figures are used as [Lighthouse's throttling default](https://github.com/GoogleChrome/lighthouse/blob/8f500e00243e07ef0a80b39334bedcc8ddc8d3d0/lighthouse-core/config/constants.js#L19-L26) and represent roughly the bottom 25% of 4G connections and top 25% of 3G connections. They are identical to the [WebPageTest "Mobile 3G - Fast" preset](https://github.com/WPO-Foundation/webpagetest/blob/master/www/settings/connectivity.ini.sample) and, due to a lower latency, slightly faster for some pages than the [WebPageTest "4G" preset](https://github.com/WPO-Foundation/webpagetest/blob/master/www/settings/connectivity.ini.sample).
These exact figures are used as [Lighthouse's throttling default](https://github.com/GoogleChrome/lighthouse/blob/8f500e00243e07ef0a80b39334bedcc8ddc8d3d0/lighthouse-core/config/constants.js#L19-L26) and represent roughly the bottom 25% of 4G connections and top 25% of 3G connections (In Lighthouse it is sometimes called "Slow 4G" used to be labeled as "Fast 3G"). This preset is identical to the [WebPageTest's "Mobile 3G - Fast"](https://github.com/WPO-Foundation/webpagetest/blob/master/www/settings/connectivity.ini.sample) and, due to a lower latency, slightly faster for some pages than the [WebPageTest "4G" preset](https://github.com/WPO-Foundation/webpagetest/blob/master/www/settings/connectivity.ini.sample).

## Throttling basics
## Types of throttling

1. Simulated throttling, which Lighthouse uses by default, is computed throttling _after a trace has been recorded_ which makes it very fast and deterministic. However, due to the imperfect nature of predicting alternate execution paths, there is inherent inaccuracy that is summarized in this doc: [Lighthouse Metric Variability and Accuracy](https://docs.google.com/document/d/1BqtL-nG53rxWOI5RO0pItSRPowZVnYJ_gBEQCJ5EeUE/edit). The TLDR: while it's roughly as accurate or better than DevTools throttling for most sites, it suffers from edge cases and a deep investigation to performance should use _Packet-level_ throttling tools.
1. The DevTools network throttling, which Lighthouse uses for `throttlingMethod='devtools'`, is implemented within Chrome at the _request-level_. As a result, it has some downsides that are now summarized in this doc: [Network Throttling & Chrome - status](https://docs.google.com/document/d/1TwWLaLAfnBfbk5_ZzpGXegPapCIfyzT4MWuZgspKUAQ/edit). The TLDR: while it's a [decent approximation](https://docs.google.com/document/d/10lfVdS1iDWCRKQXPfbxEn4Or99D64mvNlugP1AQuFlE/edit), it's not a sufficient model of a slow connection. The [multipliers used in Lighthouse](https://github.com/GoogleChrome/lighthouse/blob/3be483287a530fb560c843b7299ef9cfe91ce1cc/lighthouse-core/lib/emulation.js#L33-L39) attempt to correct for the differences.
1. _Proxy-level_ throttling tools do not affect UDP data, so they're not recommended.
1. _Packet-level_ throttling tools are able to make the most accurate network simulation.
Within web performance testing, there are four typical styles of throttling:

## How do I get high-quality packet-level throttling?
1. **_Simulated throttling_**, which Lighthouse uses by **default**, uses a simulation of a page load, based on the data observed in the initial unthrottled load. This approach which makes it both very fast and deterministic. However, due to the imperfect nature of predicting alternate execution paths, there is inherent inaccuracy that is summarized in this doc: [Lighthouse Metric Variability and Accuracy](https://docs.google.com/document/d/1BqtL-nG53rxWOI5RO0pItSRPowZVnYJ_gBEQCJ5EeUE/edit). The TLDR: while it's roughly as accurate or better than DevTools throttling for most sites, it suffers from edge cases and a deep investigation to performance should use _Packet-level_ throttling tools.
1. **_Request-level_** throttling, also referred to as _DevTools throttling_ or in the Audits Panel as _Applied throttling_, is how throttling is implemented with Chrome DevTools. In real mobile connectivity, latency affects things at the packet level rather than the request level. As a result, this throttling isn't highly accurate. It also has a few more downsides that are summarized in [Network Throttling & Chrome - status](https://docs.google.com/document/d/1TwWLaLAfnBfbk5_ZzpGXegPapCIfyzT4MWuZgspKUAQ/edit). The TLDR: while it's a [decent approximation](https://docs.google.com/document/d/10lfVdS1iDWCRKQXPfbxEn4Or99D64mvNlugP1AQuFlE/edit), it's not a sufficient model of a slow connection. The [multipliers used in Lighthouse](https://github.com/GoogleChrome/lighthouse/blob/3be483287a530fb560c843b7299ef9cfe91ce1cc/lighthouse-core/lib/emulation.js#L33-L39) attempt to correct for the differences.
1. **_Proxy-level_** throttling tools do not affect UDP data, so they're decent, but not ideal.
1. **_Packet-level_** throttling tools are able to make the most accurate network simulation. While this approach can model real network conditions most effectively, it also can introduce [more variance](https://docs.google.com/document/d/1BqtL-nG53rxWOI5RO0pItSRPowZVnYJ_gBEQCJ5EeUE/edit) than request-level or simulated throttling. [WebPageTest uses](https://github.com/WPO-Foundation/wptagent/blob/master/docs/remote_trafficshaping.md) packet-level throttling.

If you want to use more accurate throttling, read on.
Lighthouse, by default, uses simulated throttling as it provides both quick evaluation and minimized variance. However, some may want to experiment with more accurate throttling...

This Performance Calendar article, [Testing with Realistic Networking Conditions](https://calendar.perfplanet.com/2016/testing-with-realistic-networking-conditions/), has a good explanation of packet-level traffic shaping (which applies across TCP/UDP/ICMP) and recommendations.
## How do I get packet-level throttling?

### Using `comcast` for network throttling
This Performance Calendar article, [Testing with Realistic Networking Conditions](https://calendar.perfplanet.com/2016/testing-with-realistic-networking-conditions/), has a good explanation of packet-level traffic shaping (which applies across TCP/UDP/ICMP) and recommendations.

The `comcast` Go package appears to be the most usable Mac/Linux commandline app for managing your network connection. Important to note: it changes your **entire** machine's network interface. Also, **`comcast` requires `sudo`** (as all packet-level shapers do).

Expand Down

0 comments on commit 10d9c6f

Please sign in to comment.