Skip to main content
Uber logo

Schedule rides in advance

Reserve a rideReserve a ride

Schedule rides in advance

Reserve a rideReserve a ride
Engineering

Vertical CPU Scaling: Reduce Cost of Capacity and Increase Reliability

14 July 2022 / Global
Featured image for Vertical CPU Scaling: Reduce Cost of Capacity and Increase Reliability
Image
Figure 1: Key metrics for Ubers stateful management platform.
Image
Figure 2: Illustration of how a storage container has a CPU utilization (blue), a peak utilization  measured for the entire period (green), a current allocation (orange), and an optimal allocation to reach (red). The scaler gradually converges the allocation to the optimal value over a period of weeks.
Image
Figure 3: Histogram of the peak CPU utilization ratio before and after applying CPU scaling to the majority of Schemaless instances. Low is defined as a peak usage below 25%, and high as above 45%. Being in the low category is never ideal, but sometimes necessary. Being in the high category makes sense for storage instances not affected by incidents/failovers.
Image
Figure 4: The steps involved in calculating the peak CPU utilization for a given storage cluster. The default is to always look two weeks back to ensure weekend peaks are well represented in the dataset.
Lasse Vilhelmsen

Lasse Vilhelmsen

Lasse Vilhelmsen is a Staff Software Engineer at Uber where he is the Tech Lead of the Stateful Capacity and Efficiency team.

Posted by Lasse Vilhelmsen

Category: