meteor-benchapp
meteor-benchapp copied to clipboard
yauh
Benchmarking app for Meteor
Meteor Benchmarking
This is a simple benchmarking test comparing performance of Node.js and nginx.
Disclaimer These are some preliminary tests I have conducted on a weekend. I still have to prepare a more thorough test series that takes into account more real-world scenarios.
TL;DR
The test clearly shows that putting nginx in front of a Node.JS (or Meteor) application greatly increases response times for initial requests to the root path and static assets.
However, Node.JS performs much better for serving a single image than for serving the root path.
Naturally, having a multi-core system where nginx can take advantage of all cores will bring better results compared to Meteor which runs on a single core by default.
Scenario and setups
Two servers:
- meteor
- comet
Meteor runs Node.js, MongoDB, and nginx. Comet acts as a client. These tests have been conducted on a single core, single processor machine. Here are the results with 2 cores (all other specs remain).
Setup 1: Test Root for Node.JS
- Config file Node.JS
Deployment via mup, all requests go directly to Meteor at http://meteor:3000
Setup 2: Test Root for Nginx as proxy
- Config file Node.JS
- Config file Nginx
Use the configuration for nginx as a proxy. Nginx forwards all requests to the upstream Node.JS server at Port 3000.
Setup 3: Test Root for Nginx as proxy + serving static files
- Config file Node.JS
- Config file Nginx
Similar to setup 2 - proxy all requests to / to Port 3000 (Node.JS), but all requests for CSS, JS, and images are served by Nginx directly.
Setup 4: Request an image from Node.JS
- Config file Node.JS
Similar to the tests before, only this time ab requests an image (100k jpg file). Served from Node.js.
Setup 5: Request an image from Nginx as Node.JS proxy
- Config file Node.JS
- Config file Nginx
Serving the image from Nginx, with a configuration to pass requests to Node.JS.
Setup 6: Request an image from Nginx serving static
- Config file Node.JS
- Config file Nginx
Added static files as locations to Nginx config, serving from Nginx rather than Node.JS.
Results
How to read the graphs
Testing Root
Requesting the root path using Apache Bench (ab) with
ab -n100000 -c100 -g plotdata.tsv http://meteor/
| Result | Node.JS | Nginx (just proxy) | Nginx (serving static assets) |
|---|---|---|---|
| Requests | 100000 | 100000 | 100000 |
| Time taken (s) | 299.050 | 106.313 | 108.380 |
| Requests per second | 334.39 | 940.62 | 922.68 |
| Time per request (ms) (mean, across all concurrent requests) | 2.991 | 1.063 | 1.084 |
| Transfer rate (Kbytes/sec) | 269.41 | 806.51 | 791.12 |
Node.JS
Apache Bench Results | Raw Timeseries Data
Nginx as proxy
Apache Bench Results | Raw Timeseries Data
Nginx serving static assets
Apache Bench Results | Raw Timeseries Data
Testing an image
Requesting an image (100018 bytes) with
ab -n100000 -c100 -g plotdata.tsv http://meteor/image.jpg
| Result | Node.JS | Nginx (just proxy) | Nginx (serving static assets) |
|---|---|---|---|
| Requests | 100000 | 100000 | 100000 |
| Time taken (s) | 144.642 | 150.812 | 61.633 |
| Requests per second | 691.36 | 663.08 | 1622.52 |
| Time per request (ms) (mean, across all concurrent requests) | 1.446 | 1.508 | 0.616 |
| Transfer rate (Kbytes/sec) | 67719.94 | 64968.57 | 159044.71 |
Node.JS
Nginx as proxy
Apache Bench Results | Raw Timeseries Data
Nginx serving static assets
Apache Bench Results | Raw Timeseries Data
Combined results
| Result | 1a) Node.js (/) | 2a) Node.js (image.jpg) | 1b) Nginx (/, proxy) | 2b) Nginx (image.jpg, proxy) | 3a) Nginx (/, optimized) | 3b) Nginx (image.jpg, optimized) |
|---|---|---|---|---|---|---|
| Time for test | 299 | 145 | 106 | 151 | 108 | 62 |
| Reqs/s | 334 | 691 | 940 | 663 | 923 | 1623 |
| Time per req (ms) | 2.991 | 1.446 | 1.063 | 1.508 | 1.084 | 0.616 |
| Conn times min (ms) | 33 | 110 | 21 | 3 | 22 | 7 |
| Conn times mean (ms) | 299 | 145 | 106 | 151 | 108 | 62 |
| Conn times max (ms) | 490 | 294 | 260 | 473 | 163 | 1055 |
Percentage of reqs served in certain time
| Result | 1a) Node.js (/) | 2a) Node.js (image.jpg) | 1b) Nginx (/, proxy) | 2b) Nginx (image.jpg, proxy) | 3a) Nginx (/, optimized) | 3b) Nginx (image.jpg, optimized) |
|---|---|---|---|---|---|---|
| 50% | 331 | 139 | 105 | 149 | 107 | 60 |
| 66% | 346 | 151 | 110 | 167 | 111 | 63 |
| 75% | 357 | 154 | 112 | 181 | 114 | 66 |
| 80% | 364 | 156 | 113 | 191 | 116 | 67 |
| 90% | 397 | 164 | 118 | 223 | 122 | 73 |
| 95% | 420 | 175 | 124 | 255 | 127 | 80 |
| 98% | 447 | 194 | 131 | 293 | 132 | 90 |
| 99% | 459 | 206 | 134 | 315 | 135 | 98 |
| 100% | 490 | 294 | 260 | 473 | 163 | 1055 |
Specs
Both machines are equally equipped:
- Same underlying host
- KVM machines with
- 2 GB RAM
- 1 Proc with 1 Core (virtual)
- Intel E1000 NIC (virtual)
Misc
Plotting of the graphs was done using Gnuplot with inspiration from Brad Landers. Use the gnuplot script like this:
$ gnuplot gnuPlotScript
yauh