Co-operation of DiffServ and IntServ

1
Co-operation of DiffServ and IntServ
2
Co-operation of DiffServ and IntServ

• To bring advantages from both models
• Dynamic admission control and scalable core
  network
• Dynamic quantitative end-to-end services
• RSVP is used to request admission into the
  DiffServ network
• Core network is based on DiffServ
• IntServ services are mapped to the DiffServ
  codepoints at the edge router
• Network core is still statically provisioned

3
Combined IntServ/DiffServ Network
4
Measurements

• Goals
• To compare QoS of the IntServ model and the
  co-operation model
• To realize scalability problems of IntServ/RSVP
• To study DiffServ scheduling mechanisms

5
Test Network
6
Test Cases

• Three network conditions
• Case 1 unloaded, 15 real-time data (23 flows)
• Case 2 overloaded, 30 real-time data (35 flows)
• Case 3 overloaded, 45 real-time data (70 flows)
• Five network configurations
• both links are best-effort
• both links are IntServ/RSVP
• the access link is IntServ, the core link is
  DiffServ, PQ implementation
• the access link is IntServ, the core link is
  DiffServ, CBQ implementation
• the access link is IntServ, the core link is
  best-effort

7
Results for the Test Flow
8
Delay Figures
Figure 1 Case 1. BE Figure 2 Case 3. BE Figure
3 Case 3. RSVP
Figure 1.
Figure 3
Figure 2
9
Delay Figures
Figure 4 Case 3. RSVP-DiffServ (PQ) Figure 5
Case 3. RSVP-DiffServ (CBQ) Figure 6 Case 3.
RSVP-BE
Figure 4.
Figure 6.
Figure 5.
10
Conclusions

• Aggregating does not effect QoS of a microflow if
  there exists enough capacity for the aggregate
• The router we used was able to handle under 50
  RSVP reservations
• Priority queuing gave better service quality for
  the real-time flows than class-based-queuing
  (background traffic was not measured)
• more information http//atm.tut.fi/faster/