CFD Optimization of the cooling of Yosemite Open Compute Server

Abstract

Rising energy demands in data centers have constantly made thermal engineers

to think and come up with innovative cooling solutions in data centers. It is of utmost

importance to have control over the environmental impact of Data Centers. In 2011, the

Open Compute Project was started which aimed at sharing energy efficient practices for

data centers. The hardware and electrical specifications of the first open compute server –

Freedom was shared on open compute project’s website. It was a vanity free design and

its components were custom designed. It was deployed in one of the data centers in

Prineville, Oregon and within first few months of operation, considerable savings in every

and cost were observed. Since then, many open compute servers have been introduced

for applications like – compute, storage, etc. The open compute servers which were being

introduced mainly had a 2 socket architecture. Yosemite Open compute server was

introduced for serving heavy compute workloads. It provided significant improvement in

performance per watt as compared to previous generations open compute servers.

Yosemite Open Compute Server has a system on a chip architecture and has 4

CPU’s (1 CPU per sled). This study involves optimization of Yosemite Open Compute

server to improve its cooling performance using. CFD tools are very useful for thermal modeling of these servers and predict their efficiency. A commercially available CFD tool

has been used to do the thermal modeling of the server and its optimization has been done

to improve the cooling performance of the server. The model of the improved design has

been compared to the existing design to show the impact of air flow optimization on the

cooling performance of the server. The air flow characteristics and utilization of the fans

have been significantly improved in the improved design.