The performance of program execution can be described in terms of speed, or load. Typically two parameters here include:
- Latency
- Bandwidth
From DDIA (Talking about scalability)
After describing load, you can investigate what happens when a load parameter increases. The performance of your system is also something that needs to be characterized and defined.
When people talk about a p99 or a p50, this refers to percentiles. Usually performance is analyzed in the context of a p50 (or the median), but it at times, makes sense to also optimize performance for a p99 since they could be the largest user-base. A cost analysis is required to make this deduction.
Related
Be faster by omitting unnecessary work which means:
- Avoiding calculating intermediate results you don’t need
- Compute results to only the accuracy that you need in the final output
- Use caching (be prepared by knowing what the user would ask for in advanced)
- Use improved algorithms and compiler optimizations
- Spend more money to reduce hardware-bound programs (CPU-bound, swapping to disk)