Computer Science Stack Exchange is a question and answer site for students, researchers and practitioners of computer science. It only takes a minute to sign up.
Sign up to join this community
I am studying system design for distributed systems and in this page (https://github.com/donnemartin/system-design-primer), one of the following advantages was mentioned for federation for databases was:
Smaller databases result in more data that can fit in memory, which in turn results in more cache hits due to improved cache locality.
What I don't understand is why would smaller databases have more memory? Furthermore, how does federation improve cache locality and result in more cache hits?
It's not smaller databases have more memory - it's about size of a data set.
Let's say we have a data set of 100G and we have a cache of 10G - assuming completely random access to data, we have 1/10 probability of reading data from the memory cache vs reading it all the way from a storage.
What if we take our 100G and shard (e.g. using hash of a key) it to two nodes of 50G each? In that case, the probability of fining the data in the cache on every node is 10/50-> 1/5 - twice better than before.
Can we do our system even faster? Let's have 10 nodes! Now every node has just 10G of data and 10G of memory - we can read all data from the memory - making it fast.
