Dowson - The ATM Available Bit Rate Service BT/URI Distributed Systems Research Group Doc. Ref: OBU/BTURI/003
TITLE The Simulation of Available Bit Rate (ABR) Service at Different Virtual Connection Level AUTHOR Xuhui Chang, Geoff Tagg, Avril Smith, John Adams ORGANISATION Oxford Brookes University ABSTRACT DATE OF ISSUE 19th, Dec. 1995 NUMBER OF PAGES 12 ISSUE STATUE ISSUE 2: Draft for discussion CONFIDENTIALITY Confidential to the Multi-Service Networks project members, using SuperJANET DISTRIBUTION OBU Team, BT Team, Avril Smith, John Adams, Jon Crowcroft, Ian Marshall. COPYRIGHT Dec-95 Oxford Brookes University
The copyright of this document is reserved on behalf of the collaborating bodies by the originating organisation. The contents may not be copied or disclosed outside the above without permission. Permission is hearby given to JISC to reporduce this document provided that this copyright notice is included.
AUTHOR'S SIGNATURE:
__________________________________________________________________________
The Simulation of Available Bit Rate (ABR) Service
at Different Virtual Connection Level
Xu Hui Chang, Geoff Tagg Avril Smith
Oxford Brookes University University of the West of England
John Adams
BT Research Laboratories
November, 1995
1 Introduction
The Available Bite Rate (ABR) Service is intended to provide users with cell loss guarantee but without delay guarantee. It is currently under development by the ATM Forum to support traditional data applications over ATM networks. After an intense debate, the ATM Forum has finally chosen a rate-based feedback mechanism for the ABR service. A network provider can provide ABR services either at a VP level or a VC level, or both.
By implementing the ABR service at the VP level, we would like to have the following benefits:
By implementing the ABR service at the VC level, we would like to have the following benefits:
By the introduction of extensions to the ABR service, we could also:
In this paper we are going to look at three different simulations:
i) ABR PVPs with best-effort VCs (non-ABR VCs);
ii) Fixed (non-ABR PVP) PVPs, with ABR VCs,
iii) ABR at both the PVP and the VC level.
We are going to look at the issues in terms of service architecture, system performance and possible applications for these three different cases. We are also going to compare these three different cases with the network without any ABR service.
2 Introduction to Dynamic Bandwidth Controller (DBC) and Rate Controller (RC)
The dynamic bandwidth controller (DBC) and the rate controllers (RC) were proposed to provide an ABR service to existing ATM switch platforms that do not have ABR functions inside them[1].
The following is an example network under the control of DBC and RC (see Figure 1).
The DBC is attached to the edge of the network to act as a virtual source and virtual destination (VS/VD), proving explicit rate control for the segment before it. It also provides traffic buffering, policing and shaping functions.
The RC is attached to the middle switch, and its function is to monitor and calculate the bandwidth allocation for the active virtual connections at its output link.
The DBC is in contact with RCs on other switches using Resource Management (RM) cells via a special control path.
The DBC and RC can both provide the ABR service either at a VP level or a VC level, which is implementation specific.
For further details of the DBC and the RC, refer to [1].
Figure 1. The network which uses DBC and RC to provide ABR service
3 ABR PVPs with Non-ABR VCs
In this scenario, VCs are non-ABR, so the feedback mechanism does not apply at the VC level but only at the PVP level. The source cannot be controlled directly, so the end-to-end flow rates can only adjusted by the higher layer (transport layer) flow control mechanism.
3.1 System Architecture
The following is an example architecture for ABR PVP, non-ABR VC networks. The Source End Systems (SESs) and Destination End Systems (DESs) are non-ABR controlled, so the virtual channels (VCs) between SESs and DESs are non-ABR.
The main function of the ABR service at PVP level is to provide dynamic bandwidth allocation for all active PVPs according to the traffic conditions. It provides a method of maximizing the PVP capacity with very few cell losses. It also removes unwanted capacity from PVPs that are quiet or underusing their allocated capacity. In section 6.2.1, a more detailed bandwidth allocation algorithm at the VP level will be given.
Figure 2. The example network with ABR Permanent Virtual Paths,
non-ABR Virtual Channels
3.2 Possible Applications
1) LAN Emulation
In applications such as LAN interconnection, multiple workstations on a LAN can send data over a public or private ATM WAN to other workstations in other LANs. These LANs can be interconnected through VPs across the ATM backbone [2]. Because LANs are connectionless whereas ATM natively supports only connection-oriented services, a LAN emulation (LANE) service is needed to support a connectionless service over ATM. Since LANs mainly support "best-effort" service, the LAN emulation service also needs to provide a similar capability to its users. An ABR service will provide this capability [3].
2) Under-Sea Cables
For the applications which use under-sea cables as transmission media, it is expected that the network will dynamically allocate the available bandwidth among active connections. In this case, it is better to implement the ABR service at the VP level to keep the link utilization high.
3) Frame relay and X.25 applications
By the introduction of ATM to frame relay networks, the existing core network used to support frame relay could evolve to an ATM cross-connect network, with frame relay user-network interfaces [4]. The ABR service can be implemented at VP level to dynamically allocate the network available bandwidth to active VPs which will improve the QoS.
4 Fixed PVPs (non-ABR) with ABR VCs
In this case, the PVPs are fixed. They are leased line PVPs, which means that during a call period the capacity allocated to them will remain unchanged. So the ABR service does not apply to the PVP level but to the VC level. The DBC attached to the switch will detect the activities on each VC and allocate the fixed PVP capacity to the active VCs. Within a virtual path, when a virtual circuit becomes inactive, the bandwidth released is shared among other active virtual circuits. Conversely when a virtual circuit becomes active, bandwidth is reallocated among all active virtual circuits. Bandwidth allocation at the virtual path level is constant, i.e. bandwidth allocated to a VP can not be given to any other VP.
4.1 System Architecture
Figure 3 shows a three node Fixed VP/ ABR VC network. The switches are ATM cross-connect. The end systems are ATM terminal equipment capable of using ABR control. DBCs are attached to the edge of the network to monitor the activities of the VCs and to allocate the fixed VP bandwidth to active VCs.
Figure 3. The network with ABR Virtual Channels, Fixed Virtual Paths
4.2 Possible Applications
The most likely applications for this case could be :
i) Critical data transfer (e.g. defence information, air-line booking)
These applications not only need cell loss guarantees, but also demanding certain limits of cell transfer delay. As soon as such a VC goes active, it is desirable for the network to assign a certain amount of bandwidth to it.
ii) Interactive Text/Data/Image Transfer (e.g. bank Transaction, credit card verification)
5 ABR PVPs with ABR VCs
In this case, both PVPs and VCs are ABR controlled (see Figure 4), but the ABR service is only applied to VCs only at the PVP end point. Between PVP end points, the ABR service is applied to PVPs. The function of ABR at the PVP level is to provide dynamic bandwidth allocation for the active PVPs. The main function of ABR at the VC level is to monitor and detect the activities of all the VCs inside the PVP. The allocated bandwidth for each VC will be changed whenever there is activity (a VC either becomes idle or active) in one of the VC's or the allocated bandwidth for the PVP has been changed.
5.1 System Architecture
The following is an example architecture of an ABR PVP/VC network. PVP1 and PVP2 are user-to-user VPs. The network is only concerned with routing and controlling the VPs, not the individual VCs within it. So the RC only provides dynamic bandwidth allocation at the VP level. The DBC that is attached to switch 1 not only provides ABR service at the VP level, but also at the VC level.
Figure 4. The network with ABR Permanent Virtual Paths and ABR Virtual Channels
5.2 Possible Applications
The possible applications for this case could be any traditional data applications (e.g. file transfers, remote login etc.) and critical data transfer where the end-systems require a guaranteed QoS and can support the ABR service.
In order to provide a more reliable service for applications such as critical data transfer, we are going to investigate the addition of priority control bandwidth allocation policy at the VC or VP level.
6 Simulation Design
6.1 Simulation Topology
To simulate the functions of the DBC and the RC and the effects in different virtual connection levels, we used the following simple three-node model (see Fig. 5).
Two DBCs are attached to switch 1 and 2 respectively. The RC is attached to switch 2. There are two PVPs between the SESs and the DESs. Each virtual path has a number of virtual connections.
We are going to simulate the link utilization for link 2 (between switch 2 and 3) under different scenarios.
Figure 5. Simple three nodes simulation topology
6.1.1 Traffic Models
In our simulation, each source end system generates traffic according to a three-state traffic model (See figure 5).
Figure 5. The three-state traffic model
A source can be either in an ACTIVE state or an IDLE state. During the ACTIVE state, packets are generated with a pause between every packet. The packets will then be segmented into cells which will sent out to the switch. The ACTIVE state is geometrically distributed with a mean Ton. The IDLE state is an exponential distribution with mean Tidle. We chose a mean active time of 2.0s and a mean idle time of 1.0s in our simulation.
6.2 Rate Allocating Algorithm
The main disadvantages of PVCs and PVPs are removed with ABR. Bandwidth is not permanently allocated to a PVC or a PVP, when activity ceases. ABR controls remove allocated bandwidth, and the customer only gets a further allocation when next active. Because of this allocation/deallocation policy, a dynamic bandwidth allocation algorithm is required.
6.2.1 Rate allocation algorithm at PVP level
An improved algorithm is being investigated taking account of the minimum cell rate, the details of this algorithm will be supplied in our next report.
6.2.2 Rate allocation algorithm at VC level
For the ABR PVP/VC case, we not only need to allocate the available link capacity over active ABR PVP connections, but also need to allocate a PVP assigned bandwidth to the active ABR VC connections inside it.
6.2.2.1 Mean allocation algorithm (Max-min)
In this allocation algorithm, all the active VCs will share the PVP bandwidth equally. Every time when there is an activity (VC changes its state) or the network reallocate the bandwidth for the PVP, the DBC will signal the SESs with the new allocated rate. This algorithm only applies to the connections with zero-MCR.
6.2.2.2 Weighted allocation algorithm
In this algorithm, a VC has a pre-determined weight w(i) and the bandwidth allocation for the VC is proportional to w(i). W(i) can be determined by any one of the following:
( i ) Proportional to the MCR of the connection,
( ii ) By application type, with different applications having different weights. For example, assume inside a PVP, there are three VCs which are for telnet, ftp and mail respectively, a possible weighting might be:
w(1)=0.4 - for telnet;
w(2)=0.5 - for ftp;
w(3)=0.1 - for mail.
How to categorize applications into different types with different weights is for further study,
( iii ) Proportional to the burstiness of the connection.
6.2.3 Summary
This bandwidth allocation algorithm fulfills the requirements of an allocation/deallocation policy, guaranteeing minimum cell rate (MCR) to VPs, giving a fair share and keeping link utilization high. It removes bandwidth from VPs which no longer require their current allocation.
6.3 Simulation Results
Simulation Duration: 5 seconds,
Link Bandwidth: 10 mbits/s (23584 cells/s),
Number of PVPs: 2,
Link trunk delay: 0.001 second,
Buffer size inside the DBC for each PVP: 35 cells,
For figures 6-9, 6 VCs are set up, 3 VCs for each PVP, the initial cell rate(ICR) and MCR for each source is 1.5 Mbits/s and 0 Mbits/s respectively. For figure 10-11, 9 VCs are set up, 3 VCs for PVP1, 6 VCs for PVP2 , the ICR is 1.0 Mbits/s, the MCR is 0 Mbits/s.
Figure 6. Link utilization for Fixed PVP/ABR VC
Figure 7. Link utilization for Fixed PVP/ABR VC with smaller time granularity
Figure 6 and Figure 7 show the simulation results for the case of fixed PVP and ABR VC with different time granularity. From Figure 6 we can see that the link utilization could increased by 30 % with the ABR control at VC level than without any ABR control under the same simulation environment. The cell loss is zero. The problem for this case is that when there is no VCs active inside a VP, the bandwidth allocated to that VP will be wasted. The link utilization will drop.
Figure 8 and Figure 9 show the simulation results for the case of ABR PVP and ABR VC with different time granularity. With the control of DBC and RC at VP and VC level, we achieved around 30% higher link utilization without cell losses than without any control.
Figure 8. Link utilization for ABR PVP/VC
Figure 9. Link utilization for ABR PVP/VC with smaller time granularity
Figure 10. Cell loss ratio for ABR PVP/non-ABR VC
Figure 11. Link utilization ABR PVP/non-ABR VC
Figure 10 and Figure 11 show the results of the case that ABR control at VP level but not VC level. In this simulation we use 9 VC sources rather than 6, 3 VCs in PVP1, 6 VCs in PVP2. The ICR is 1.0 Mbits/s for each source. For the non-ABR VP and non-ABR VC case, the PVPs equally share the link bandwidth, but the PVP with ABR control will get dynamically allocated bandwidth from the network according to its current load. The results show that with the ABR control at the VP level, we achieved higher link utilization as well as low cell loss ratio.
7 Conclusions and Work for the Future
In this paper, we have carried out research for the ABR service operating at different virtual connection level. The simulation results show that the DBC and RC are promising for incorporating ABR service into existing ATM switches not only at PVP level but also at VC level.
In the future, we are going to carry on our simulation studies to investigate different bandwidth allocation algorithms at both the VP and the VC level. The simulation for different traffic sources should also be carried out. We also plan to undertake some simulations over a more complicated topology (with more nodes, where the network will segmented into several control loops, we are going to investigate the DBC's VS/VD functions over it and also fairness over the active connections). It is also very interesting to do some research on the delay effect over the whole network performance (e.g. stability, transient performance etc.).
References
[1] J L Adams and A J Smith, "The support of the Available Bit Rate bearer capability using Virtual Source/Destination concepts", BT Technology Journal Vol.13 No.3 July 1995
[2] Shirish S. Sathaye, "Traffic Management Specification Version 4.0", ATM Forum/95-0013R8, Straw Vote, October 1995.
[3] K.-Y.Siu and R.Jain, "A Brief Overview of ATM: Protocol Layers, LAN Emulation, and Traffic management", Computer Communication Review, pp6-20, Vol. 25, No. 2, April, 1995.
[4] L. G. Cuthbert and J-C Sapanel, " ATM- The Broadband Telecommunications Solution", The Institution of Electrical Engineers, pp87.
[5] Brook N.C, "High Level Design for Rate-Controlled Dynamic Bandwidth Controller Simulation Model", DSN1/MSNP/212/WD(ABR) Draft 2A, BT Labs, Martlesham Heath, Ipswich, 1995.