1. Assuring QoS Guarantees for
Heterogeneous Services in
RINA Networks with ΔQ
Sergio Leon(1), Jordi Perelló(1), Davide Careglio(1), Eduard Grasa(2),
Miquel Tarzan(2), Neil Davies(3) and Peter Thompson (3)
(1) Universitat Politècnica de Catalunya (UPC)
(2) Fundació Privada i2CAT (i2CAT)
(3) Predictable Network Solutions
* This research has been funded by the European Project FP7 PRISTINE, as well as the
Spanish National project SUNSET.

2. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
Summary
 Applications and their requirements
 A quick look at RINA
 QoS in RINA
 RINA and ΔQ scheduling policies
 Scenario description
 Numerical results
 Conclusions
2

3. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
Applications and their requirements
 Application requirements have evolved over time:
 Limits on delay and acceptable jitter
 Possibility and distribution of acceptable losses
 Secure flows
 Etc.
 IP with TCP/UDP does not fulfil current requirements
 Other solutions based on IP neither!
 Next generation networks should be able to get and
fulfil QoS requirements at each layer
 Here comes RINA
3

4. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
A quick look at RINA(1/2)
 RINA : Recursive InterNetworking Architecture
 Key idea:
“Networking is Inter Process
Communication (IPC) and only IPC”
 What it really is:
 Clean-slate recursive Internet model
 Same type of layer, Distributed IPC Facility (DIF), at each level
 All DIFs share the same functionality
 Even so, each DIF can be fully configured via policies
 Each DIF may provide full support for a wide range of QoS
4

5. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
A quick look at RINA(2/2)
 Each IPC Process (IPCP) shares the same API
 Upper IPCP and applications use that to request flows with
specific QoS requirements
5

6. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
QoS in RINA
 Multiple QoS Cubes available at each DIF
 QoS Cubes define measures like:
 Minimum quality provided:
• Maximum delay and jitter
• Probability of loss (0 if reliable)
• Minimum bandwidth
 Maximum quality provided:
• Maximum bandwidth
• Maximum supported burst and duration
 Additional capabilities:
• Flow security
• Cost and assurance on overbooked scenarios
6

7. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
RINA and ΔQ scheduling policies(1/4)
 Scheduling is the main way to ensure short-term QoS
requirements
 In particular, when performing scheduling, 3 measures
become tightly linked:
Delay / Losses / Bandwidth
Three parameters: only two degrees of freedom!
 In RINA, QoS Cubes provide simple bandwidth constraints
 And requirements in terms of delay and losses
 Just what we need for ΔQ-aware scheduling policies
7

8. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
RINA and ΔQ scheduling policies(2/4)
 ΔQ scheduling policies focus on sharing unavoidable
degradation
 How much added delay a packet/flow suffers?
 How probable is it that packets are dropped?
 What will the throughput of a flow be?
 ΔQ scheduling policies merge:
 Multiplexing (Cherish/Urgency matrix)
 Flow shaping Cherish 1 Cherish 2 Cherish 3
Urgency 1 A1 A2 A3
Urgency 2 B1 B2 B3
Urgency 3 C1 C2 C3
8

9. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
RINA and ΔQ scheduling policies(3/4)
 ΔQ manages inter-flow contention
 Cherish/Urgency multiplexor per output port
 Shared buffer between all flows
 Capacity to discard an incoming packet given the flow cherish
level and global occupation
 Serve based on urgency
 Shares degradation
between flows given
their QoS
9

10. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
RINA and ΔQ scheduling policies(4/4)
 ΔQ manages intra-flow contention
 Policer/Shaper per flow, QoS, set of flows, etc.
 Shaping long bursts and spacing packets avoids the
starvation of low requirement flows
 Policing ensures
that flows behave
as contracted
10

11. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
Scenario Description(1/2)
 Inter-datacenter communication between multiple
network domains
 QoS has to be supported at each layer
 Particularly in backbone networks
11

12. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
Scenario Description(2/2)
 Our focus is a small backbone network of 10 nodes
 It supports traffic
 Internet traffic: Best-effort traffic (BE) and Urgent traffic (sBE)
 Inter-DC traffic: Urgent traffic (GU) and non-urgent (SN)
 Inter-DC traffic is more valuable, but with same urgencies
12

13. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
Numerical results(1/2)
 ΔQ policies allows us to ensure
minimum or even null losses
for high-priority traffic even at
150% link capacity
 Latency can be also contained
for urgent traffic,
independently of their losses
13
C/U + Cherish ··· - Cherish
+ Urgent GU sBE
- Urgent SN BE

14. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
Numerical results(2/2)
 Compared to existing
IP-based solutions:
 Losses can be gracefully
shared between QoS
classes. as well as latency
and jitter
 This means that:
 Degradation is not
equally shared between
flows
 Nor does it concentrate in
the lowest priority classes
only!
14
Weighted Fair QueueBaseline Best-Effort

15. UPC
Sergio Leon, Jordi Perelló, Davide Careglio, Eduard Grasa, Miquel Tarzan, Neil Davies and Peter Thompson,
“Assuring QoS Guarantees for Heterogeneous Services in RINA Networks with ΔQ”
Conclusions
 Assuring QoS guarantees is a requirement for future
networks
 Not only that, but a way to inform and maintain these
requirements between layers
 RINA provides a model capable of fully supporting QoS
for current and future application requirements
 In this regard, ΔQ scheduling policies provide a way to
effectively ensure QoS guarantees, even in overloaded
scenarios
15

16. Thanks. Questions?
The authors of this work would like to thank all members
of the PRISTINE Project consortium for the valuable
discussions and inputs