Skip to main content
Version: 4.6.1

Statistics Best Practices

This guide explains how to analyze statistics data and make practical decisions. It covers practical topics such as how to set Limited Inflow and how to plan capacity adjustments.

Regular Monitoring Checklist​

Weekly Review:

  1. Check Outflow Rate (should be 80% or higher)
  2. Check Queue Size and Wait Time during peak hours
  3. Compare Process Time between normal hours and peak hours

Monthly Review:

  1. Review peak Inflow patterns over the past month
  2. Analyze capacity utilization (identify periods when Queue/Wait Time increased)
  3. Check integration health (identify segments with consistently low Outflow Rate)

When Planning Capacity Changes:

  1. Analyze 3-6 months of historical data (use Month view)
  2. Identify peak periods and check server resources (refer to APM records)
  3. Monitor for 1-2 weeks after changes (use Day view)

Determining Optimal Limited Inflow​

Step 1: Understand Normal and Peak Periods​

What to check:

  1. Check Inflow: Check the rate of initial requests (Entry Requests) coming in during normal hours and peak hours
  2. Check queue conditions: Check how many users are waiting (Queue Size) and average wait time (Wait Time) at those times

Example Pattern:

Time    Inflow (TPS)    Queue Size    Wait Time    Limited Inflow    Interpretation
09:00   80              20            3s           100              Normal hours
10:00   120             50            8s           100              Peak starting
11:00   150             200           20s          100              Peak (waiting occurs)
12:00   130             180           18s          100              Peak continues

Step 2: Evaluate Limited Inflow Appropriateness During Peak​

Evaluation Criteria:

  1. Check server resources: Check WAS server CPU and other computing resource usage during peak hours by referring to APM records
  2. Decision:
    • Server has available capacity but wait time is long → Consider increasing Limited Inflow
    • Server is overloaded and wait time is long → Maintain or decrease Limited Inflow
    • Server has available capacity and wait time is low → Keep current settings

What to check:

  • Queue Size and Wait Time during peak hours
  • Server CPU usage during peak hours (refer to APM records)
  • Whether the server has available capacity or is overloaded

Example Evaluation:

Peak hour situation:
- Inflow: 150 TPS
- Queue Size: 200 users
- Wait Time: 20 seconds
- Server CPU usage: 50% (APM records)

Decision: Server has available capacity but wait time is long → Consider increasing Limited Inflow

Step 3: Check Performance Degradation via Process Time​

Important Principle:

  • Ideal situation: For Basic Segments, normal hour processing time (Process Time) and peak hour processing time should be nearly identical.

Pattern Analysis:

  1. Are normal hour and peak hour Process Time similar?

    • Similar → Normal, adjust Limited Inflow based on Queue/Wait Time and server resources

  2. Has Process Time increased only during peak hours?

    • Increased → Server may be responding slowly
    • If server resources are still available (check APM records), consider increasing Limited Inflow even if queue size increases
    • If server is overloaded, do not increase Limited Inflow; investigate and fix the performance issue first

Example Pattern:

Time    Process Time    Queue Size    Interpretation
09:00   2.5s            20            Normal hours (normal)
10:00   2.6s            50            Peak starting (normal)
11:00   4.5s            200           Peak - Process Time increased (server response delay)
12:00   4.2s            180           Peak continues - Process Time increased

Decision: Process Time increased only during peak hours → Possible server response delay
→ If server resources are available (check APM), consider increasing Limited Inflow even if queue size increases

Action:

  • If Process Time during peak hours has increased significantly compared to normal hours, it may indicate a server performance issue
  • First, check server resources (CPU, memory via APM records):
    • If server has available capacity: Consider increasing Limited Inflow to allow more concurrent requests, which may help if the delay is due to queuing rather than server overload
    • If server is overloaded: Do not increase Limited Inflow; investigate and resolve the performance bottleneck first
  • Simultaneously investigate server response delay causes through server logs or APM

Detecting Integration Issues​

Pattern: Inflow vs Outflow Divergence​

What it means:

  • Inflow consistently higher than Outflow = Users are entering but not completing service properly
  • Low Outflow Rate (<80%) = Missing nfStop() calls or integration problems

Example Pattern:

Time    Inflow (TPS)    Outflow (TPS)    Outflow Rate    Interpretation
09:00   100             95               95%             Healthy
10:00   100             80               80%             Healthy
11:00   100             60               60%             Issue - missing exit calls
12:00   100             55               55%             Issue - code review needed

Action:

  1. Immediate: Check which segments have low Outflow Rate (use Segment view)
  2. Temporary measure: If keys are not being returned, adjust key return timeout to force automatic key return
    • Lower the timeout (e.g., minimum 6s → 3-4s) to make automatic return happen faster
    • This won't be reflected in Outflow Rate, but capacity will be released faster so other users can enter
    • Timeout settings can be adjusted in segment Advanced - Timing (Basic Control) or Advanced - Timing (Section Control)
  3. Root cause investigation: Review recent code changes, find missing nfStop() calls
  4. Root cause fix: Add explicit exits to all code paths, ensure error handling includes key return

Important Considerations When Changing Capacity​

Post-Change Monitoring​

After changing Limited Inflow, always monitor:

  1. Immediately after change: Monitor for 1-2 weeks using Day view
  2. What to check:
    • Whether Queue/Wait Time has improved
    • Whether expected effects have appeared
  3. Adjust: Make additional adjustments based on actual results

Incremental Change Principles​

  • Increase: Increase by 10-20% at a time, monitor, then repeat
  • Decrease: If server protection is urgent, decrease by 40-50% immediately; otherwise, decrease gradually