Twitter has designed a new search architecture that took the company's search gurus about six months to build. The search team built the search architecture on a more efficient inverted index rather than a relational database. They chose to start with Lucene, a search engine library written in Java.
More than 1,000 tweets per sec and 12,000 queries per sec that equal more than 1 billion queries per day puts a heavy load on machines, so the new system must last for several years with the ability to scale.
The search team estimates the platform only uses about 5% of the available backend resources, which means there's room to grow. The new index tool can potentially process roughly 50 times more Tweets per second than it does today.
Read more about it on the company's blog.