Peak to Trough

The importance of auto-scaling 

peak to trough traffic

The cloud enables us to acquire hardware on demand for our services. I have never had to rack a server or worry about hardware failure. My entire software career has been in the cloud. As an industry most of us don’t need to worry about forecasting hardware requirements months in advance. We just increase the number of virtual machines we need in the PAAS dashboard. 

This week I was investigating some unusually large peaks in our daily traffic. I was changing the bounds and timeline of the graph and noticed that we had a 10x difference peak to trough. Usage peaks for about 2 hours each day at 10x trough, about 6 hours are also peak but at 5x trough. At night our traffic drops significantly because our users are sleeping. 

My current team, like all teams I have worked with in my five year career, does not use auto-scaling. We experimented with it last year but had issues with auto-scaling interfering with our deployments in unpredictable ways. 

So we scale for our instantaneous peak of 10x our lowest traffic around 2am. Meaning we use at least 5x as much hardware as necessary. 

The drawing underestimates the impact of the instantaneous peaks which essentially double the traffic to this service. 

Auto-Scaling would be a great fit for this service. Most cloud platforms have supported this use case for years and would result in decent savings. 

If you like my writing, please buy my book on Amazon.
The Sledgeworx Guide to Getting into Software