Google uses a "wrench" for redundancy on 100,000 servers

Google's global infrastructure has launched a proprietary system that automatically transfers and repeat loads when large data volumes are unwilling to communicate with the network switching load in case of hardware problems.

This distributed technology was first unveiled at a conference named "Google classic fashion" (classically coy Google fashion) this summer, google Academician Jeff Dean confirmed the existence of this technology at a seminar earlier this month.

This platform is called "Spanner ?). In Dean's presentation, the platform was described as follows: "storage and computing systems cover Automatic Data Center movement, enhanced data replication and computing usage restrictions and patterns ." Includes bandwidth, packet loss, resource restrictions, energy consumption, and "failure mode ".

Dean is talking about "automatic allocation of a whole column of machine resources"-Google now has at least 36 large data centers around the world, and some may still be under construction. As mentioned earlier, Google's new system is hoping to span a large data center fleet.

According to Dean's speech, Google hopes that the Spanner can control 1 million to 10 million servers, including 10 trillion (1013) directories and 1000 trillion (1018) bytes of storage space. All of this is distributed across data centers around the world.

Imagine that an independent big house is using cables to control other data centers in the world.

 

Dean declined to comment. Google's PR department did not provide a specific response to this issue, but Vijay Gill, Senior Manager of Google's engineering and architecture department, mentioned the technology at a mini Conference previously held in San Francisco.

When asked, "If you can wave your magic wand to create a backend network technology," Gill said, "we do not have this technology now, "when talking about Google's famous distributed online infrastructure, he was a little mysterious-Google turned the data center into a" warehouse scale "machine, when a data center is overloaded, it is moved to another place.

"What we need to do now is-of course, this is a warehouse-sized computer," Gill says, you must have all the rights from cool-down to CPU consolidation ."

"Sometimes, there is a temperature change, and you may need a fast load switch to organize the temperature change. Is there a chiller in your data center? You want to reduce the load and reduce the number of CPU and RAM processes ."

He said that the company can perform automatic or near-automatic intervention without manual intervention. "How do you optimize global management systems? This is an interesting phenomenon ."

"As we can see, Google's Large-Scale Linear Programming Problems involve 100,000 variables, almost all of which require real-time computing. When the temperature in a data center changes, you don't have time to set the temperature in other data centers. You have to make a judgment within several seconds ."

When asked whether Google is using the technology, Gill replied that this is just what Google is most interested in. "I cannot make a comment," he said. "I don't remember making any file ."

But it seems that the technology described by Gill is about Spanner. According to Academician Dean's speech, it seems that the technology has been deployed. Google also said that it was not running in a new data center in Saint Ghislain, Belgium. Apparently, Spanner technology was used to make it possible to spend the hot summer.

Dean said that Spanner aims to provide channels for data transmission within 50 subtle points. In addition, Google has the opportunity to deploy at least two storage devices in Europe for storage, two in the United States, and one in Asia.

Obviously, Google has the talent for distributed computing.

1. What kind of "Internet tsunami" Will Google's death bring"
2. Google secrets: Zero bandwidth fee for high traffic
3. Google Chrome browser exposure