Bandwidth Management
Content
Intra-domain Bandwidth Management in Differentiated Services Network
S. Jha1, M. Hassan, P. Nanda and N. Ahmed Network Research Laboratory School of Computer Science and Engineering The University of New South Wales, Australia E-mail: [email protected]
Abstract
In absence of any link layer traffic controls or priorityqueuing mechanism in LAN infrastructure (such as shared media LAN), Subnet Bandwidth Management based approach of managing bandwidth is limited to only total amount of traffic load imposed by RSVP associated flows. In such cases no mechanism is available to separate RSVP flows from Best Effort Traffic. This brings usefulness of subnet Bandwidth Manager into question. This paper attempts to use a combination of Integrated Services of Specific Link layer model based on RSVP and IP ratecontrol approach for best effort traffic to manage intradomain traffic in a Differentiated services network.
provide QoS guarantees in Intradomain environment in Diffserv networks.
2. Subnet Bandwidth Manager Background
Subnet bandwidth manager (SBM) as defined in the Internet draft [3] is a signaling protocol for LAN based admission control supporting RSVP to map into IEEE 802 style networks. To Map Integrated Services such as controlled load and guaranteed service over IEEE802 style LAN, SBM and ISSLL use "aggregated flows". Each flow is assigned to one of the priority classes. The admission control and class identification modules in switches ensure that the service requirement for traffic in each class is met. In this context, each Diffserv domain may consist of several managed segments. A single SBM acts as a designated SBM (DSBM) and is responsible for admission control over the resource reservation requests originating from the source hosts. The SBM [3] proposes an extension to RSVP for managing resources of a shared segment. Initially the host sends it’s RSVP PATH message through the DSBM. The DSBM forwards this message to the next hop (router/other IntServ device) and waits for the RSVP RESV message to be back from the destination. DSBM keeps track of all resources consumed on the segment for all installed reservations and determines whether or not to accept subsequent reservations based on remaining resources. However this approach works only for RSVP managed traffic. Since there is no control over best effort traffic, reservation made by RSVP flows is meaningless.
1. Introduction
Future Internet environments will provide support for Quality of Service (QoS) management, which will enable applications, such as teleteaching, video on demand, and virtual laboratories that are not possible with the technology underlying today's Internet. The current Internet provides best effort service with no mechanisms for providing guarantees about the loss or delay experienced by the data packets. This is currently being addressed by Integrated Services (IntServ) [1] and Differentiated Services (DiffServ) [2] initiatives within the IETF. In general it is accepted that, the future Internet will consist of IntServ access networks, and DiffServ backbone network. Integrated Services over Specific Link Layer (ISSLL) proposes use of Subnet Bandwidth Manager (SBM) to support QoS in shared media environment [3]. In absence of any link layer traffic control or priority-queuing mechanism in LAN infrastructure (such as shared media LAN), SBM based approach of managing bandwidth is limited to only total amount of traffic load imposed by RSVP associated flows. In such cases no mechanism is available to separate RSVP flows from Best Effort (BE) traffic. This brings usefulness of subnet Bandwidth Manager (SBM) into question. It may be possible that BE receives better quality than EF, because EF is subject to rate control but BE is not. This paper proposes a new approach by combining IntServ and IP-Rate control methods to
3. IP Rate Control Scheme
The Bandwidth Broker/egress router periodically (at T seconds interval) calculates the required aggregate input rate R to the egress router which will keep the queue length q at a pre-specified threshold Q. This rate R is applicable to best effort traffic only. In order to model the egress router buffer, we use linear feedback control [4] system using the following equation
R = C + K (Q − q )
Where K is the control gain and C is the current available link rate. K is feed forwarded to reduce the impact of noise on control system. The bandwidth broker/Egress Router maintains a variable N for the number of hosts competing for link bandwidth, which is obtained from the number of hosts which transmitted one or more packet through the egress router in last W seconds. At the beginning of each interval (T seconds), the Egress router calculates (or obtains from Bandwidth broker) a single value for Explicit Rate (ER) as ER = max (0, R )
Packet Scheduler based on rate notification message (RSVP_RN) manages these queues. Thus the scheduler ensures that packets leaving the host fit their negotiated (EF traffic) and prescribed (BE traffic).
5. Experimental results
Some preliminary experimental results based on a simple configuration have shown promising results. Experimental setup is with two 10Mbps Ethernet segments connected via two egress routers with a serial link between them using Point to Point Protocol (PPP). The serial link is configured to 115Kbs (approx. 128Kbps ISDN link connecting corporations in WAN environment). Hosts are configured using Linux that uses standard TCP/IP stack as well as modified stack to support IP rate control. Figure 2 shows the max. queue length at the egress router. This result clearly shows that maximum queue length is very small in comparison to the standard TCP/IP implementation. Average queuing delay also remains stable (and very low) as the number of connections increases for the IP rate controlled Best Effort traffic.
100
N This ER value is the fair share of available bandwidth for all hosts. Once the ER is computed, the next task for the gateway is to notify hosts of the new ER. The rate-controlled work described in this paper extends the Host Gateway Rate Control Protocol [4]. Section 4 proposes an integrated scheme by extending RSVP to include new message types rate and queue notification.
4. Implementation management
model
for bandwidth
Figure 1 shows our proposed scheme which extends the SBM to manage Best Effort traffic as well. For a possible implementation of this model at individual hosts, RSVP
PATH RESV
Max. Queue Length (Packets)
90 80 70 60 50 40 30 20 10 0 1
IP Rate control Standard
BANDWIDTH BROKER
PATH PATH RESV RESV
Number of TCP Connections
2
3
4
Link to other domain
Figure 2 Max. Queue Length
DSBM RSVP_RN RSVP_QN RSVP_RN RSVP_QN
6. Conclusions and Future Work
We have proposed a hybrid model that uses a combination of ISSLL based on RSVP and IP rate control approach for Best Effort traffic to manage intra-domain traffic in a Differentiated services network. Future work will implement the proposed model and benchmark efficiency.
Figure 1 DiffServ Implementation Model
daemon needs to support new RSVP messages RSVP_RN and RSVP_QN as described above. The egress router sends the ER values to the DSBM in a new type of RSVP message for rate notification (RSVP_RN). DSBM broadcasts (or multicasts as proposed in [3]) this to its managed LAN. Egress router maintains separate queue for each outgoing link (if they are connected to multiple domains). A queue notification (RSVP_QN) is sent from the egress router to hosts to indicate that it should use a particular queue for this router (one queue for each possible outgoing link). In addition to marking DSCP for EF traffic based on DCLASS objects and putting them in separate set of queues, Packet Classifiers at host puts BE packets in queues identified by RSVP_QN messages. The
7. References
[1] R. Braden, D. Clark, and S. Shenker, "Integrated Services in the Internet Architecture: an Overview", IETF RFC 1633, Jun. 1994. [2] S. Blake, D. Blake, M. Carlson, E. Davies, Z. Wang, W. Weiss,"An Architecture for Differentiated Services", IETF RFC 2475, Dec. 1998. [3] R. Yavatkar, D. Hoffman, Y. Bernet, F. Backer, M. Speer, "SBM (Subnet Bandwidth Manager): A Protocol for RSVP-based Admission control over IEEE 802-style networks", IETF RFC 2814, May 2000. [4] M. Biggs, M. Hassan and J. Breen, "Implementation and Experimentation of IP Rate Control", Proceedings of IEEE International Conference on Networking, ICON'99, Sep. 1999.
S. Jha1, M. Hassan, P. Nanda and N. Ahmed Network Research Laboratory School of Computer Science and Engineering The University of New South Wales, Australia E-mail: [email protected]
Abstract
In absence of any link layer traffic controls or priorityqueuing mechanism in LAN infrastructure (such as shared media LAN), Subnet Bandwidth Management based approach of managing bandwidth is limited to only total amount of traffic load imposed by RSVP associated flows. In such cases no mechanism is available to separate RSVP flows from Best Effort Traffic. This brings usefulness of subnet Bandwidth Manager into question. This paper attempts to use a combination of Integrated Services of Specific Link layer model based on RSVP and IP ratecontrol approach for best effort traffic to manage intradomain traffic in a Differentiated services network.
provide QoS guarantees in Intradomain environment in Diffserv networks.
2. Subnet Bandwidth Manager Background
Subnet bandwidth manager (SBM) as defined in the Internet draft [3] is a signaling protocol for LAN based admission control supporting RSVP to map into IEEE 802 style networks. To Map Integrated Services such as controlled load and guaranteed service over IEEE802 style LAN, SBM and ISSLL use "aggregated flows". Each flow is assigned to one of the priority classes. The admission control and class identification modules in switches ensure that the service requirement for traffic in each class is met. In this context, each Diffserv domain may consist of several managed segments. A single SBM acts as a designated SBM (DSBM) and is responsible for admission control over the resource reservation requests originating from the source hosts. The SBM [3] proposes an extension to RSVP for managing resources of a shared segment. Initially the host sends it’s RSVP PATH message through the DSBM. The DSBM forwards this message to the next hop (router/other IntServ device) and waits for the RSVP RESV message to be back from the destination. DSBM keeps track of all resources consumed on the segment for all installed reservations and determines whether or not to accept subsequent reservations based on remaining resources. However this approach works only for RSVP managed traffic. Since there is no control over best effort traffic, reservation made by RSVP flows is meaningless.
1. Introduction
Future Internet environments will provide support for Quality of Service (QoS) management, which will enable applications, such as teleteaching, video on demand, and virtual laboratories that are not possible with the technology underlying today's Internet. The current Internet provides best effort service with no mechanisms for providing guarantees about the loss or delay experienced by the data packets. This is currently being addressed by Integrated Services (IntServ) [1] and Differentiated Services (DiffServ) [2] initiatives within the IETF. In general it is accepted that, the future Internet will consist of IntServ access networks, and DiffServ backbone network. Integrated Services over Specific Link Layer (ISSLL) proposes use of Subnet Bandwidth Manager (SBM) to support QoS in shared media environment [3]. In absence of any link layer traffic control or priority-queuing mechanism in LAN infrastructure (such as shared media LAN), SBM based approach of managing bandwidth is limited to only total amount of traffic load imposed by RSVP associated flows. In such cases no mechanism is available to separate RSVP flows from Best Effort (BE) traffic. This brings usefulness of subnet Bandwidth Manager (SBM) into question. It may be possible that BE receives better quality than EF, because EF is subject to rate control but BE is not. This paper proposes a new approach by combining IntServ and IP-Rate control methods to
3. IP Rate Control Scheme
The Bandwidth Broker/egress router periodically (at T seconds interval) calculates the required aggregate input rate R to the egress router which will keep the queue length q at a pre-specified threshold Q. This rate R is applicable to best effort traffic only. In order to model the egress router buffer, we use linear feedback control [4] system using the following equation
R = C + K (Q − q )
Where K is the control gain and C is the current available link rate. K is feed forwarded to reduce the impact of noise on control system. The bandwidth broker/Egress Router maintains a variable N for the number of hosts competing for link bandwidth, which is obtained from the number of hosts which transmitted one or more packet through the egress router in last W seconds. At the beginning of each interval (T seconds), the Egress router calculates (or obtains from Bandwidth broker) a single value for Explicit Rate (ER) as ER = max (0, R )
Packet Scheduler based on rate notification message (RSVP_RN) manages these queues. Thus the scheduler ensures that packets leaving the host fit their negotiated (EF traffic) and prescribed (BE traffic).
5. Experimental results
Some preliminary experimental results based on a simple configuration have shown promising results. Experimental setup is with two 10Mbps Ethernet segments connected via two egress routers with a serial link between them using Point to Point Protocol (PPP). The serial link is configured to 115Kbs (approx. 128Kbps ISDN link connecting corporations in WAN environment). Hosts are configured using Linux that uses standard TCP/IP stack as well as modified stack to support IP rate control. Figure 2 shows the max. queue length at the egress router. This result clearly shows that maximum queue length is very small in comparison to the standard TCP/IP implementation. Average queuing delay also remains stable (and very low) as the number of connections increases for the IP rate controlled Best Effort traffic.
100
N This ER value is the fair share of available bandwidth for all hosts. Once the ER is computed, the next task for the gateway is to notify hosts of the new ER. The rate-controlled work described in this paper extends the Host Gateway Rate Control Protocol [4]. Section 4 proposes an integrated scheme by extending RSVP to include new message types rate and queue notification.
4. Implementation management
model
for bandwidth
Figure 1 shows our proposed scheme which extends the SBM to manage Best Effort traffic as well. For a possible implementation of this model at individual hosts, RSVP
PATH RESV
Max. Queue Length (Packets)
90 80 70 60 50 40 30 20 10 0 1
IP Rate control Standard
BANDWIDTH BROKER
PATH PATH RESV RESV
Number of TCP Connections
2
3
4
Link to other domain
Figure 2 Max. Queue Length
DSBM RSVP_RN RSVP_QN RSVP_RN RSVP_QN
6. Conclusions and Future Work
We have proposed a hybrid model that uses a combination of ISSLL based on RSVP and IP rate control approach for Best Effort traffic to manage intra-domain traffic in a Differentiated services network. Future work will implement the proposed model and benchmark efficiency.
Figure 1 DiffServ Implementation Model
daemon needs to support new RSVP messages RSVP_RN and RSVP_QN as described above. The egress router sends the ER values to the DSBM in a new type of RSVP message for rate notification (RSVP_RN). DSBM broadcasts (or multicasts as proposed in [3]) this to its managed LAN. Egress router maintains separate queue for each outgoing link (if they are connected to multiple domains). A queue notification (RSVP_QN) is sent from the egress router to hosts to indicate that it should use a particular queue for this router (one queue for each possible outgoing link). In addition to marking DSCP for EF traffic based on DCLASS objects and putting them in separate set of queues, Packet Classifiers at host puts BE packets in queues identified by RSVP_QN messages. The
7. References
[1] R. Braden, D. Clark, and S. Shenker, "Integrated Services in the Internet Architecture: an Overview", IETF RFC 1633, Jun. 1994. [2] S. Blake, D. Blake, M. Carlson, E. Davies, Z. Wang, W. Weiss,"An Architecture for Differentiated Services", IETF RFC 2475, Dec. 1998. [3] R. Yavatkar, D. Hoffman, Y. Bernet, F. Backer, M. Speer, "SBM (Subnet Bandwidth Manager): A Protocol for RSVP-based Admission control over IEEE 802-style networks", IETF RFC 2814, May 2000. [4] M. Biggs, M. Hassan and J. Breen, "Implementation and Experimentation of IP Rate Control", Proceedings of IEEE International Conference on Networking, ICON'99, Sep. 1999.
