It talks about how LinkedIn optimized the feature of showing 1st,2nd,3rd+ degrees on any profile we search on application. In their distributed graph DB, to solve this problem efficiently, they're doing online computation to calculate 2nd degree connections for each member and storing in cache(NCS). But in case of millions of active users, low latency is much needed, hence to keep it, they used NP-Hard greedy set cover algorithm to only make calls to minimal no. of nodes to compute result faster.