Knowledgebase

Understanding Latency in Integrations Print

  • 0

Where time goes.

WHAT CONTRIBUTES

Network round trip The provider's processing Your own processing Serialisation Waiting on connections

WHAT DOMINATES OVER DISTANCE

The round trip.

WHY THAT MATTERS

An operation making ten sequential calls pays ten round trips.

WHAT TO DO ABOUT IT

Reduce the number of calls Make independent calls concurrently Move work closer

WHAT CONCURRENCY ACHIEVES

Several round trips overlapping rather than adding.

WHAT IT REQUIRES

Calls genuinely independent of each other.

WHAT TO BE CAREFUL WITH

Concurrency against rate limits Error handling across several simultaneous calls

WHAT TO MEASURE

Latency at percentiles, per provider and per operation.

WHY PERCENTILES

The slow tail is what users experience as broken.

WHAT TO SET FROM THAT MEASUREMENT

Timeouts.

WHAT TO CONSIDER ABOUT CHAINS

Total latency is the sum, and total reliability is the product.

WHY THAT SECOND POINT MATTERS

Five services at ninety-nine per cent each give well under ninety-nine overall.

WHAT THAT ARGUES FOR

Fewer synchronous dependencies.

WHAT TO MOVE OUT OF THE REQUEST PATH

Anything not needed to answer the user.


Was this answer helpful?
Back

Are you happy with your experience? Leave us a review on Trustpilot.


Trustpilot