Transcript Packet Context-aware Queue Management

Autonomic QoS Management Mechanism
in Software Defined Network
Speaker: Chang, Cheng-Yu
Advisor : Dr. Kai-Wei Ke
Date: 09/Dec./2014
Outline
•
•
•
•
•
•
Introduction
Quality of Service
Software Defined Network
Autonomic QoS Management Mechanism in SDN (AQSDN)
Packet Context-aware QoS model (PCaQoS)
Experimental Evaluation
• Conclusion
• References
2
Introduction
• As the increment of various network equipments and services, the
complexity of network control and management has risen sharply in recent
years.
• The QoS management is an important part of network management.
• Various QoS models and mechanisms have been proposed, but there is
no one-fit-all algorithm.
• The self-configurable QoS models and mechanism based on the contextaware are highly expected.
3
Introduction (Cont.)
• In order to reduce the network management cost and the probability of
network failure.
• Both Software Defined Network(SDN) and Autonomic Network
technologies are sophisticated technologies for the network control and
management.
• Design an Autonomic QoS management mechanism in SDN for network
QoS guarantee.
• In this mechanism, the controller undertakes the function of analysis and
decision in the autonomic control loop.
4
Quality of Service
• The goal of QoS is to provide guarantees on the ability of a network to
deliver predictable results.
• Elements of network performance within the scope of QoS often include
•
•
•
•
availability (uptime)
bandwidth (throughput)
latency (delay)
error rate
5
Quality of Service (Cont.)
1. Mark and Classify packets according to policies and the behavior of the
traffic. This is carried out with technologies such as IP Precedence and
DSCP and is most effective.
2. Congestion management by prioritising traffic based on the marks using
queuing technologies that can respond to traffic classes.
3. Avoid congestion by dropping packets that are not a high priority using
technologies such as Random Early Detection.
4. Limit the ingress or egress traffic depending on the class/markings of the
packets. Also perform traffic shaping to maximise the use of bandwidth
by specifying peak and average traffic rates.
5. Fragment and compress packets to maximise the use of WAN
bandwidths.
6
Quality of Service (Cont.)
Important QoS technologies / protocols:
• Class of Service [802.1p/Q] - layer 2
• Integrated Services (IntServ) - layer 3
• Differentiated Services (Diffserv) - layer 3
7
Class of Service [802.1p/Q]
• Layer 2 Class of Service can be provided within the TCI field of the
Ethernet frame
Priority
Services
0 (最低)
Routine
1
Priority
2
Immediate
3
Flash
4
Flash Override
5
Critical
6
Internetwork Control
7 (最高)
Network Control
8
Integrated Services (IntServ)
•
•
•
•
Manages traffic on a per-flow basis
Provides customized services per traffic stream
End-to-end application registration
Resource Reservation Protocol (RSVP)
9
Differentiated Services (Diffserv)
•
•
•
•
Manages traffic on a type-of-traffic basis
Does not provide individual stream visibility
Implemented per hop
Type of Service(ToS), DiffServ Code Point (DSCP)
10
SDN Architecture
11
AQSDN Architecture
• The QoS control module
decides or chooses QoS rules
dynamically.
12
QoS control module
Is an application of the controller, has two function models:
• QoS scheme decision model: determines suitable queue management
and scheduling scheme as well as their parameters. (infrequently)
• QoS action decision model: determines packets marking and designates
the queue for each adaptively. (frequently)
QoS schemes and actions collectively referred as QoS rules.
13
QoS control module: DB
• Requirement DB: stores the QoS requirements
• Rule DB: stores the historical QoS rules
• Policy DB: stores the QoS Polices supported by OpenFlow switch
14
QoS control module: Context manager
In QoS management include network context and flow context
• Network Context:
• State information (e.g., utilization ratio of CPU, length of Packet queue)
• Link information (e.g, Packet loss ratio delay, jitter, bandwidth of link)
• Flow Context:
• Inherent feature (e.g., service type, QoS requirement)
• Real-time flow featue (e.g., burst rate)
15
QoS control module: Analysis
Analyze whether the QoS requests can be satisfied and if there are conflicts
among them.
• If conflicts are founded or the QoS request could not be satisfied:
-> Sent to the administrator
• If the QoS requests could be satisfied:
-> Forwarded contexts and requirements to the QoS rule decision module
16
QoS control module: Rule decision
Can chooses the appropriate QoS rule from the QoS Rule DB.
• If Rule DB not exits: Plans new QoS rule(include context, requirement,
policy) and stores into the Rule DB.
17
QoS action
• The packet marking algorithms selection is configured via OpenFlow
protocol by OpenFlow controller.
• The meter table and meter band which is defines by OpenFlow protocol
provide the meter operation for flow.
• The existing remarking bands remarking bands, which only lower the drop
precedence level of the packet, does not satisfy various QoS requirement.
• It is necessary to extend OpenFlow protocol so that it could support
diverse packet marking algorithms.
18
QoS action (Cont.)
• We explore the specific arguments instead of the predefinition of the structure for
each meter band.
• We add two packet marking algorithms as new meter bands, Single Rate Three
Color Marker(srTCM) and Packet Context-aware Packet Marker(PCaPM)
OpenFlow 1.3.1: Switch Hardware (Forwarding Plane)
19
QoS scheme
• For each queue, queue management, schedule schemes are configured
through OF-Config protocol.
• The existing OF-Config has provided the support of the minimum and
maximum transmission rates of a queue.
• For supporting the configuration of Queue management and Queue
scheduling schemes, we enrich the operate set of the OF-Config on the
content layer.
20
QoS scheme (Cont.)
• PQ: Priority Queue
• WRR: Weighted Round-Robin
• WRED: Weighted Random Early Detection
21
Packet Context-aware QoS Model (PCaQoS)
• Process packets according to their semantic precedence level.
• The QoS guarantee ability would be improved if the SDN take packet
context into account.
• It is impractical to deliver the packet context to the controller because this
needs the frequent communication between the switch and the controller.
• Design the PCaQoS enhanced from the DiffServ model, which enables the
switch to perceive the packet context and responds it locally.
22
PCaQoS (Cont.)
• In addition to the metering information about the flows, the marker and the
queue manager also take the packet contexts into account.
• They called as Packet Context-aware Packet Marker (PCaPM) and
Packet Context-aware Queue Management (PCaQM)
logical view of packet classification and traffic
23
Packet Context-aware Packet Marker (PCaPM)
• PCaPM initiate a multicolor packet marker by remarking the DSCP code
based on the packet context and the marking result of the metering-based
marker.
• The metering-based marker is srTCM, the packets of a service present
three kinds of priority (high, middle and low) and the mapping relationships
from meter result to remarking color.
• Single Rate Three Color Marker(srTCM) marks packets as either Green,
Yellow, or Red. [RFC 2698]
24
PCaPM (Cont.)
25
Packet Context-aware Queue Management
• In a congested route, if lots of lower priority packets arrive suddenly in a
short time, the queue length increases sharply so that the heiger priority
packets coming later will be dropped.
• So PCaQM takes packet priority as another metric for the icoming packets
processing.
26
PCaQM (Cont.)
• PCaQM derives k virtual sub queues from the original queue. (k is colors
or priorites)
• All of packets with the same priority l(1≤l≤k) in original queue Q compose
sub queue sq[l].
27
PCaQM Algorithm
PROCEDURE PCaQM (Packet p, Queue Q) {
l:= p.PL;
qlen:= av_len(Q);
Pr:= Calc_Discard_Probability(qlen,
Q.thmin[l], Q.thmax[l], Q.pmax[l]);
if ( Packet_Should_be_Discarded (Pr) ) {
ReplaceDiscard(p, Q);
}
else {
Q. AppendPacket(p) ;
Q.sq[l]. AppendPacket (prep) ;
}
}
PROCEDURE ReplaceDiscard (Packet p, Queue Q) {
l:= p.PL;
sqlow := null;
for (i:= k to l+1 step -1) do
if (Q.sq[i] == null) continue;
sqlow := Q.sq[i] ;
break;
if (sq low == null)
Discard(p);
else
p rep := sq low. .GetQueueHeader ();
Q.RemovePacket(prep) ;
sqlow.RemovePacket (prep) ;
Q. AppendPacket(p) ;
Q.sq[l]. AppendPacket (prep) ;
}
28
Experimental Evaluation
The prototype system of AQSDN, which config QoS polices in SDN according to the policy
of multimedia service autonomically.
•
•
•
•
•
Controller: Dell R710 run NOX platform
Switches: run Ofsoftswitch13 support both DiffServ and PCaQoS
Bandwidth between CS1 and CS2 is 20Mbit/s
Bandwidth of any other link is 100Mbit/s
PCs serve as source and destination hosts.
The implementation of prototype system
29
The self-configuration feature of the AQSDN
• In order to compare PCaQoS and DiffServ, we drive two virtual network.
First virtual network
Second virtual network
Switches
ES1, CS1, CS2
ES2, CS1, CS2, ES4
Flow classification
and
Queue scheduling
Vedio flows: AF class (PQ+WRR)
Vedio flows: AF class (PQ+WRR)
Background flows: AF and BF(WRR) Background flows: AF and BF(WRR)
class
class
Packet marker
And
Queue management
ES1: srTCM
Other: WRED
ES2: PCaPM
Other: PCaQM
Referred to as
DiffServ based network
PCaQoS based network
30
PCaQoS vs. DiffServ
• Present the average Peek Signal to Noise Ratio(PSNR) of each video after
across PCaQoS and DiffServ based network respectively.
31
PCaQoS vs. DiffServ (Cont.)
• Select 300 continuous frames from three videos after across the PCaQoS
and DiffServ based network respectively, and analyze their PSNR.
32
Conclusion
• In the traditional IP network, resource utilization improvement and network
QoS guaranteeing are very complicated for network operators.
• Propose them is to upgrade the network nodes with autonomic abilities.
• SDN provides the capability to implement network control and
management functions by software
33
Conclusion (Cont.)
• The AQSDN architecture, which combines the advantages of the
autonomic network management and the SDN technologies.
• A novel QoS model which is called PCaQoS model is also presented
based on the AQSDN architecture.
• The self-configuration feature of the AQSDN and the enhancement of
video quality of the PCaQoS model are verified.
34
Reference
• Wang Wendong. “Autonomic QoS Management Mechanism in Software
Defined Network.” China Communications, vol.11, pp13-23. July 2014.
35
Thanks for Listening