Home » QoS Deep Dive: DiffServ, DSCP, Queuing, Policing, Shaping และ Voice/Video QoS
QoS Deep Dive: DiffServ, DSCP, Queuing, Policing, Shaping และ Voice/Video QoS
QoS Deep Dive: DiffServ, DSCP, Queuing, Policing, Shaping และ Voice/Video QoS
QoS (Quality of Service) จัดลำดับความสำคัญของ network traffic เพื่อให้ critical applications ทำงานได้ดีแม้ network congested DiffServ เป็น QoS architecture ที่ classify + mark traffic ด้วย DSCP, Queuing จัดการลำดับการส่ง packets, Policing จำกัด traffic rate โดย drop excess, Shaping จำกัด traffic rate โดย buffer excess และ Voice/Video QoS ต้องการ specific treatment เพื่อ quality ที่ดี
เมื่อ network congested (traffic > capacity) ทุก traffic ได้รับผลกระทบเท่ากัน ถ้าไม่มี QoS: VoIP call ขาดๆ หายๆ, video conference กระตุก, database replication ช้า ในขณะที่ backup traffic หรือ software updates ใช้ bandwidth เท่ากัน QoS แก้ปัญหาโดยให้ critical traffic ได้ priority สูงกว่า
QoS Architecture: DiffServ
| Step |
Action |
Where |
| 1. Classification |
ระบุ traffic ว่าเป็นประเภทไหน (voice, video, data, best-effort) |
Access layer (edge) |
| 2. Marking |
Mark packet ด้วย DSCP value ใน IP header |
Access layer (edge) |
| 3. Queuing |
จัด traffic เข้า queues ตาม DSCP → schedule ส่งตาม priority |
Every hop |
| 4. Policing |
จำกัด traffic rate (drop excess traffic ที่เกิน limit) |
Edge / WAN interface |
| 5. Shaping |
จำกัด traffic rate (buffer excess → send later) |
WAN interface (outbound) |
DSCP Values
| DSCP |
Per-Hop Behavior |
Decimal |
Traffic Type |
| EF |
Expedited Forwarding |
46 |
Voice (low latency, low jitter, low loss) |
| AF41 |
Assured Forwarding 41 |
34 |
Video conferencing (interactive video) |
| AF31 |
Assured Forwarding 31 |
26 |
Streaming video |
| CS6 |
Class Selector 6 |
48 |
Network control (routing protocols: OSPF, BGP) |
| CS5 |
Class Selector 5 |
40 |
Signaling (VoIP signaling: SIP, H.323) |
| AF21 |
Assured Forwarding 21 |
18 |
Transactional data (ERP, CRM, database) |
| AF11 |
Assured Forwarding 11 |
10 |
Bulk data (email, FTP, backup) |
| CS1 |
Class Selector 1 |
8 |
Scavenger (P2P, non-business) |
| DF (CS0) |
Default Forwarding |
0 |
Best effort (everything else) |
Queuing Mechanisms
| Mechanism |
How |
Use Case |
| FIFO |
First In, First Out (no priority) |
Default (no QoS) — ไม่เหมาะกับ congested links |
| Priority Queuing (PQ) |
Strict priority — high queue always first |
Voice queue (EF) — ส่งก่อนเสมอ (แต่ starve queues อื่น) |
| CBWFQ |
Class-Based Weighted Fair Queuing — allocate bandwidth per class |
Guarantee bandwidth per traffic class (video 30%, data 50%) |
| LLQ |
Low Latency Queuing = PQ + CBWFQ |
Best practice: PQ for voice + CBWFQ for other classes |
| WRED |
Weighted Random Early Detection — drop before queue full |
Prevent tail drop (TCP synchronization) — drop lower priority first |
Policing vs Shaping
| Feature |
Policing |
Shaping |
| Action on Excess |
Drop (or re-mark) excess traffic |
Buffer excess → send later (delay) |
| Direction |
Inbound or outbound |
Outbound only |
| Traffic Pattern |
Bursty (allows bursts up to Bc) |
Smooth (shapes to configured rate) |
| Use Case |
ISP limiting customer bandwidth, ingress rate limiting |
Match WAN link speed, smooth traffic for downstream device |
| CPU Impact |
Low (just drop/mark) |
Higher (buffering + scheduling) |
Voice QoS Requirements
| Metric |
Requirement |
QoS Treatment |
| Latency (one-way) |
< 150 ms |
Priority queue (LLQ) — send voice first |
| Jitter |
< 30 ms |
Priority queue eliminates queuing jitter |
| Packet Loss |
< 1% |
Priority queue + bandwidth reservation |
| Bandwidth per Call |
~100 Kbps (G.711) / ~30 Kbps (G.729) |
Allocate enough bandwidth in PQ |
| DSCP Marking |
EF (46) for voice media |
Mark at phone/access switch |
| Signaling |
CS5 (40) for SIP/H.323 |
Separate queue from media |
Video QoS Requirements
| Type |
Latency |
Jitter |
Loss |
DSCP |
| Interactive Video (conferencing) |
< 200 ms |
< 30 ms |
< 1% |
AF41 (34) |
| Streaming Video |
< 4-5 sec (buffered) |
Tolerant |
< 5% |
AF31 (26) |
| Telepresence |
< 150 ms |
< 15 ms |
< 0.05% |
CS4 (32) |
QoS Design Model (Cisco)
| Queue |
Traffic |
DSCP |
Bandwidth |
| Priority Queue |
Voice |
EF |
10% (strict priority) |
| Queue 2 |
Interactive Video |
AF41 |
23% |
| Queue 3 |
Signaling + Network Control |
CS5, CS6 |
5% |
| Queue 4 |
Transactional Data |
AF21 |
18% |
| Queue 5 |
Bulk Data |
AF11 |
9% |
| Queue 6 |
Scavenger |
CS1 |
1% |
| Default Queue |
Best Effort |
DF (0) |
34% (remaining) |
ทิ้งท้าย: QoS = Right Traffic Gets Right Treatment
QoS Deep Dive DiffServ: classify → mark (DSCP) → queue → police/shape (per-hop behavior) DSCP: EF (voice), AF41 (video), CS6 (control), AF21 (transactional), DF (best effort) Queuing: LLQ = priority queue (voice) + CBWFQ (guaranteed bandwidth per class) Policing: drop excess (inbound/outbound) | Shaping: buffer excess (outbound, smooth) Voice: < 150ms latency, < 30ms jitter, < 1% loss → EF in priority queue Video: < 200ms interactive, buffered streaming → AF41/AF31 in bandwidth queue Design: mark at edge, queue at every hop, police/shape at WAN interfaces
อ่านเพิ่มเติมเกี่ยวกับ Network Performance Testing iPerf และ VPN Technologies IPsec WireGuard ที่ siamlancard.com หรือจาก icafeforex.com และ siam2r.com
