Mura is unevenness in workload over time: the spikes and troughs in production demand. Peaks can overload a system (muri), while troughs leave capacity idle (muda). section 12.5
Average load can hide the problem: production systems fail at peaks, not averages. Examine demand over time to see whether a service is designed for its actual traffic pattern. section 12.5.1
Uneven ingestion demand
An analytics platform receives 60% of its uploads between 8:00 and 9:30 a.m. The morning surge strains memory and grows queues; later, the service sits at 15% utilization. Daily volume is manageable, but its uneven distribution creates both overload and idle capacity. section 12.5.1
Level demand to reduce the effects of mura:
- For batch work, accept jobs promptly and process them as capacity becomes available, spreading work over time. section 12.5.1
- For real-time traffic, use measures such as per-customer rate limits or geographic load distribution. section 12.5.1
Table 12.5 Mura symptoms in production systems
|
Symptom
|
What it signals
|
Where to look
|
|
Periodic latency spikes at the same time daily
|
Traffic pattern mura
|
Request rate by hour of day
|
|
Auto-scaler constantly adding/removing instances
|
Demand oscillation
|
Scaling event logs, traffic histograms
|
|
Queue depth that alternates between 0 and 5,000
|
Batch job mura
|
Queue depth over time, job scheduling
|
|
On-call pages clustered on specific days
|
Operational mura
|
Incident frequency by day of week
|
|
End-of-month processing failures
|
Calendar-driven mura
|
Resource utilization by calendar week
|
Use these production symptoms to help locate uneven demand in request rates, scaling events, queues, incidents, or calendar patterns. section 12.5.1
