Systems Design Specific Tech

Question 1

In Dynamo what is the difference between Global and Local indexes?

Answer 1

They are both Secondary indexes.

Global: Both partition id and sort key are different from base table values. Stored in its own partition space and scales separately from base table.

Local: Only sort key is different from base table values. Scoped to base table partition

Question 2

Since data stores that write to disk can also keep data in memory, why are in memory data stores faster?

Answer 2

Because they don’t have to encode the data into the format it lives on disk.

Question 3

What argument can you use for just using Postgres

Answer 3

With various extensions it supports
- TimeSeries
- Machine Learning
- OLAP
- Derived
- GeoSpatial
- Search
- Graph
- Federated queries to other data stores.

Question 4

What type of replication does a Redis cluster use?

Answer 4

Leader/Follower

Question 5

How many TPS can Redis handle?

Answer 5

A good number to use for interviews
- AWS Elasticache supports 1 million TPS

Enterprise versions with 26 AWS instances perf tested at 200 million TPS.

Question 6

How many writes per second can SQL handle?

Answer 6

AWS Aurora 200,000 per second.

Question 7

What are key scalability constraints for Kafka?

Answer 7

1. Aim for less than 1 mb per message. This is an explicit constraint in Kinesis.
2. 1 TB of storage per broker
3. 1 kb message about 100,000 per second

Question 8

What are constraints for throughput and querying on SQL database?

Answer 8

MySql: 10 to 50 million writes per second.
AWS Aurora: 200 million writes per second.
Query/Joins: 10 Million records in table is going to need special indexing. 50 to 200 million, you need to consider partitioning.

Question 9

How many writes per second can DynamoDB handle?