The official Twitter account released a year's project summary . Highlights include:
-
Completely restructured the For you service and ranking system, reducing the number of lines of code from 700K to 70K, a 90% reduction, reducing computing usage by 50%, and increasing post throughput by 80%.
-
The API middleware layer of the technology stack was restructured and the architecture was simplified , removing more than 100,000 lines of code and thousands of unused internal ends, and eliminating unused client services.
-
Closed the Sacramento data center and reconfigured 5,200 racks and 148,000 servers, saving more than $100 million annually.
-
Post-metadata retrieval delays were reduced by 50%, and global API timeout errors were reduced by 90%.
-
Optimized use of cloud service providers to do more work locally, reducing monthly cloud costs by 60%. Specifically, all media/blob artifacts are moved out of the cloud, reducing the overall cloud data storage size by 60%. At the same time, cloud data processing costs have been reduced by 75%.
-
Build a local GPU supercomputing cluster and design, develop and deliver a 43.2Tbps new network fabric architecture to support the cluster.
-
Expand network backbone capacity and redundancy, saving $13.9 million annually.
Recommended reading
- An engineer modified the configuration and caused Twitter to crash. Musk responded: This pile of "ShitCode" must be completely rewritten.
- Musk hardcore migrates Twitter servers
- Musk said Twitter data was being scraped to the extreme and urgently launched a "current limiting" mechanism
- Musk ordered Twitter engineers overnight to modify the algorithm