AWS gaming experiment

I got inspired by this blog post by Larry Gadea and decided to ask about it on twitter:

So that’s what I decided to do. Here’s a story of my experiments and findings:

The traffic from Finland to Amazon AWS regions in EU seems to change a bit depending on traffic. The following snippet is from 25.4.2015 but earlier when I tested I got almost double the latency.

Cloudping to AWS

Let’s try anyway.

I decided to go with regular instance instead of spot instance due to faster start-up and simpler pricing model. First problem I encountered was that I just created a new account and Amazon needs to verify the account before I can create instances.

For testing I created a new GPU instance in Ireland and configured VPN, Steam etc. roughly following Larry’s instructions.

I then tested to play some games over the connection. I tested with the following games:

  • Cities: Skylines
  • Death Rally
  • Portal

And here’s some stats from playing (or attempting to play) Cities: Skylines.

	"SlowNetworkPercent"		"83.889091491699219"
	"SlowDecodePercent"		"0.48104065656661987"
	"SlowDisplayPercent"		"30.016672134399414"
	"AvgClientBitrate"		"147.28001403808594"
	"StdDevClientBitrate"		"137.56430053710937"
	"AvgServerBitrate"		"2135.413330078125"
	"StdDevServerBitrate"		"8527.41015625"
	"AvgLinkBandwidth"		"7525.86865234375"
	"AvgPingMS"		"1354.215087890625"
	"StdDevPingMS"		"2320.548095703125"
	"AvgCaptureMS"		"2.3584384918212891"
	"StdDevCaptureMS"		"1.7173653841018677"
	"AvgConvertMS"		"0.021369790658354759"
	"StdDevConvertMS"		"0.1436113715171814"
	"AvgEncodeMS"		"5.5488224029541016"
	"StdDevEncodeMS"		"1.5767149925231934"
	"AvgNetworkMS"		"191.907958984375"
	"StdDevNetworkMS"		"450.26995849609375"
	"AvgDecodeMS"		"2.8170154094696045"
	"StdDevDecodeMS"		"31.469053268432617"
	"AvgDisplayMS"		"38141.1484375"
	"StdDevDisplayMS"		"337954.03125"
	"AvgFrameMS"		"109.8192138671875"
	"StdDevFrameMS"		"118.69873809814453"
	"AvgFPS"		"55.433006286621094"
	"StdDevFPS"		"16.267507553100586"

I’m not sure how reliable the stats are, I noticed quite large differences in “AvgPingMS” value between different attempts and games. The range I saw was from about 200 ms to about 1400 ms.

In the end I decided that this was simply not going to work. The games were completely unplayable.

This was a fun experiment that cost me about 5USD in EC2 instance and storage costs.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.