Zipf's law.
also: Zipf distribution · ZipfianThe nth most popular item gets about 1/n of the top item's traffic, so a few keys take most requests.
the two-minute refresher
- Rank 1: 1000 req/s, rank 2: 500, rank 3: 333, rank 4: 250.
- Real traffic is skewed, not uniform: plan for the head of the curve.
- Good news for caching: a small cache holds the hottest items and gets a high hit ratio.
- Bad news for partitioning: the top keys can overload their shards.