Google Data Center

This video describes about the Google data center located at Dalles, OR, where even most of the Google employees can’t get in due to security clearance. Site Reliability Engineering Team writes and maintains the software system to maintain perfect running of the services and avoid failures. It has 24/7 team of SREs on call to tackle any problems related to unexpected failures though highly redundant power, networking and serving domain and thus preventing the loss of cluster and minimizing the impact.

The entry into the data center floor is protected by biometric iris scanner and circle lock door and would require dual authentication to get in. The data center floor showing lot of servers forming a single cluster. Management of these servers would require several tools such as Borg, Colossus and Spanner along with Kubernetes, Google Cloud Storage and Big Query by Google engineers and Cloud customers. One of the network engineer explains working of Hardware Ops to expand the data center to deploy additional machines inside the building. And Google had preplanned for all expansions of this quantity using Jupiter, which is their data center technology providing hierarchical design using software defined networking principles. The Single building supporting 75000 machines require lot of optical cable fibers and carry over one petabit per second of bandwidth. This has helped Google to reliably access storage with low latency and high throughput. It is also noted that Google runs on B4, which is their private network and has been growing faster than our Internet-facing network. This connects all of the Google data centers and allows smooth access of resources across them. Next, the Google data center’s safety protocol of data storage is explained. On a daily basis the hard drives are shredded and wiped out to protect our data when there is hard drive and SSD failure. But the protocol calls for strict chain of custody from the time of hard drive being removed and commissioned for a new one.

Next comes the important part where cooling and powering the infrastructure is explained in the video, since lot of heat are generated inside the server area and has to be removed. The cooling plant on the site has two water loops, the condenser and process water loops (differentiated by different colors). The process water loop take the heat off the server floor and the condenser water loop takes the cold water from the basin to the heat exchangers. Power usage efficiency (PUE) in most of the Google data centers were 100% with very low Power overhead. It also uses a chiller which at times required to be used to keep the water in the cooling tower at the desired temperature. It is this water which is used in condenser water loop.  This cooling tower would normally use evaporation to cool the water at a faster rate. Finally the Google-owned power station is shown, which powers the Cloud. This is where high voltage power enter the site before going to distribution centers. And also uses multiple generators to prevent outage. Best part is all the power is coming from nearby hydroelectric power station and thus achieving 100%carbon neutral.  

Reference: https://www.youtube.com/watch?v=zDAYZU4A3w0