System capacity

Estimate the system capacity

Determine if the system is skewed towards reads or writes.

Compute the read/write ratio, for example 100:1.

We need to estimate how much data storage is needed and decide how the system e.g. database scales to support reading or writing more data.

Scaling a read-heavy system makes use of replication and caching. Replication means having a copy of the same data on multiple nodes to increase the read throughput and availability if some nodes go down. Caching means keeping a copy of the most frequently read data in memory to minimise latency and bandwidth.

Scaling a write-heavy system uses partitioning (sharding). Partitioning means splitting the data across multiple nodes when it becomes too large to fit in a single one.

Replication and partitioning have implications on whether the system can offer strong or eventual consistency.

The system is read-heavy

The system is write-heavy

The system is read/write balanced

Estimate the network traffic, in number of read/write requests per unit of time.

Number of requests per month

Number of requests per second

Estimate the network bandwidth requirements.

The average size of a request in KB/s

Estimate the disk storage requirements

Decide how much time the data is expected to be stored

Estimate the total storage for the expected duration in MB

Estimate the memory requirements for caching frequently read data

Decide a percentage of data to cache, for example 20% of data generates 80% of traffic

Calculate the size of the cache

Create a summary table with these estimates

Name	Value	What does this mean?
Number of read req/s
Number of write req/s
Network bandwidth estimates (KB/s)
Persistent storage estimates (MB)
Memory cache estimates (MB)