Home » Network Performance: Latency, Jitter, Packet Loss, Throughput, QoE และ Performance Monitoring
Network Performance: Latency, Jitter, Packet Loss, Throughput, QoE และ Performance Monitoring
Network Performance: Latency, Jitter, Packet Loss, Throughput, QoE และ Performance Monitoring
Network Performance วัดคุณภาพของ network ด้วยหลาย metrics Latency คือ delay ในการส่ง packet, Jitter คือ variation ของ latency, Packet Loss คือ packets ที่หายระหว่างทาง, Throughput คือ actual data rate ที่ใช้งานได้, QoE (Quality of Experience) วัดจากมุมมองของผู้ใช้ และ Performance Monitoring ติดตาม metrics เหล่านี้อย่างต่อเนื่อง
Network ที่ “ใช้ได้” ไม่ได้หมายความว่า performance ดี: ping ได้แต่ VoIP เสียงขาด (jitter > 30ms), website เปิดได้แต่ช้า (latency 200ms+), file download สำเร็จแต่ช้ามาก (packet loss 2% → TCP throughput ลดลง 50%+) Performance monitoring ต้องวัดอย่างต่อเนื่อง ไม่ใช่แค่ตอนมีปัญหา — proactive monitoring ป้องกัน outage ก่อนเกิด
Key Performance Metrics
| Metric |
Definition |
Acceptable |
Impact |
| Latency (Delay) |
Time for packet to travel from source to destination (one-way or RTT) |
< 100ms (general), < 150ms (VoIP) |
Slow app response, poor VoIP, video buffering |
| Jitter |
Variation in latency between consecutive packets |
< 30ms (VoIP), < 50ms (video) |
Audio gaps, video pixelation, choppy calls |
| Packet Loss |
Percentage of packets that don’t reach destination |
< 1% (general), < 0.5% (VoIP) |
TCP retransmissions (slow), UDP gaps (quality) |
| Throughput |
Actual data transfer rate achieved (vs bandwidth = max capacity) |
70-90% of bandwidth |
Slow downloads, buffering, timeouts |
| Bandwidth |
Maximum theoretical data rate of link |
Utilization < 70% sustained |
Congestion, queuing, increased latency |
Latency Components
| Component |
Cause |
Typical |
| Propagation Delay |
Speed of light through fiber/copper — distance dependent |
~5ms per 1,000km (fiber) |
| Serialization Delay |
Time to put bits on wire — depends on link speed and packet size |
Negligible on 1G+ links |
| Queuing Delay |
Wait in queue at congested interface — variable, depends on traffic |
0-100ms+ (main source of variable latency) |
| Processing Delay |
Router/switch processes packet (routing lookup, ACL check, NAT) |
< 1ms per hop (modern hardware) |
| Application Delay |
Server processing time (database query, API call) |
Variable (10ms – seconds) |
Impact of Packet Loss on TCP
| Packet Loss |
TCP Throughput Impact |
User Experience |
| 0% |
Full throughput |
Normal |
| 0.1% |
~90% of max |
Barely noticeable |
| 1% |
~50-70% of max |
Noticeably slower downloads |
| 2% |
~30-50% of max |
Slow, frustrating |
| 5% |
~10-20% of max |
Nearly unusable for file transfers |
| 10%+ |
< 5% of max |
Unusable — constant retransmissions, timeouts |
QoE (Quality of Experience)
| Application |
Latency |
Jitter |
Packet Loss |
Bandwidth |
| VoIP (G.711) |
< 150ms (one-way) |
< 30ms |
< 1% |
80Kbps per call |
| Video Conference (HD) |
< 200ms |
< 50ms |
< 1% |
2-4 Mbps |
| Video Streaming (4K) |
< 500ms (buffered) |
N/A (buffered) |
< 0.1% |
25+ Mbps |
| Web Browsing |
< 200ms (page load) |
N/A |
< 1% |
5-25 Mbps |
| Online Gaming |
< 50ms (competitive) |
< 20ms |
< 0.5% |
1-5 Mbps |
| File Transfer |
Tolerant (TCP adjusts) |
N/A |
< 2% |
As much as available |
Performance Monitoring Tools
| Tool |
What It Monitors |
Type |
| SNMP (v2c/v3) |
Interface utilization, errors, CPU, memory |
Polling (pull) |
| NetFlow/sFlow/IPFIX |
Traffic flows: who talks to whom, how much, which apps |
Flow export (push) |
| IP SLA (Cisco) |
Synthetic tests: ICMP echo, UDP jitter, HTTP, DNS |
Active probing |
| ThousandEyes |
End-to-end path visibility: Internet, SaaS, cloud |
Cloud-based probing |
| PRTG |
All-in-one: SNMP, flow, packet, cloud, availability |
Multi-method |
| Grafana + Prometheus |
Custom metrics, dashboards, alerting |
Open-source, flexible |
| Wireshark |
Deep packet inspection, protocol analysis, troubleshooting |
Packet capture |
Performance Optimization
| Problem |
Solution |
| High Latency |
QoS (prioritize sensitive traffic), reduce hops, CDN, optimize routing, WAN optimization |
| High Jitter |
QoS (jitter buffer, priority queuing), upgrade congested links, traffic shaping |
| Packet Loss |
Fix errors (CRC, duplex mismatch), upgrade bandwidth, QoS (WRED for TCP), fix bad cables |
| Low Throughput |
Increase bandwidth, TCP tuning (window size), WAN optimization, reduce packet loss |
| Congestion |
Upgrade bandwidth, QoS (classify + queue), traffic shaping, load balancing (ECMP) |
ทิ้งท้าย: Performance = Measure, Monitor, Optimize
Network Performance Metrics: latency (< 100ms), jitter (< 30ms VoIP), packet loss (< 1%), throughput (70-90% of bandwidth) Latency: propagation + serialization + queuing (main variable) + processing + application Packet Loss: 1% = 50% TCP throughput reduction | 5% = nearly unusable QoE: VoIP (< 150ms, < 30ms jitter), video (< 200ms), gaming (< 50ms), web (< 200ms) Monitoring: SNMP (utilization), NetFlow (traffic), IP SLA (synthetic), ThousandEyes (end-to-end) Optimization: QoS, bandwidth upgrade, WAN optimization, TCP tuning, fix physical errors Key: measure continuously, set baselines, alert on deviations — proactive monitoring prevents outages
อ่านเพิ่มเติมเกี่ยวกับ QoS Quality of Service DiffServ DSCP และ Network Troubleshooting Methodology OSI Wireshark ที่ siamlancard.com หรือจาก icafeforex.com และ siam2r.com
