Content uploaded by Pham Thanh Giang
Author content
All content in this area was uploaded by Pham Thanh Giang on Dec 26, 2013
Content may be subject to copyright.
Contention Window Size Control for QoS
Support in Multi-Hop Wireless Ad Hoc
Networks
Pham Thanh Giang and Kenji Nakagawa
Department of Electrical Engineering, Nagaoka University of Technology,
Kamitomioka 1603–1, Nagaoka-Shi, Niigata, 940–2188 Japan,
{giang@kashiwa, nakagawa@}.nagaokaut.ac.jp
Abstract. IEEE 802.11 MAC protocol for medium access control in
wireless Local Area Networks (LANs) is the de facto standard for wireless
ad hoc networks, but it does not necessarily satisfy QoS requirement by
users such as fairness among flows or throughput of flows. In this paper,
we propose a new scheme aiming to solve per-flow fairness problem and
achieves good throughput performance in IEEE 802.11 multi-hop ad hoc
network.
We apply a cross-layer scheme, which works on the MAC and link layers.
In the link layer, we examine each flow and choose suitable contention
window for each output packet from the flow. The aim of the mecha-
nism is achieving per-flow fairness. In our proposed scheme, the MAC
layer of each station tries to evaluate network condition by itself then ad-
just channel access frequency to improve network utilization and ensure
per-node fairness. Performance of our proposed scheme is examined on
various multi-hop network topologies by using Network Simulator (NS-
2).
Keywords: cross-layer, per-flow/per-node fairness, bandwidth utiliza-
tion, multi-hop wireless, IEEE 802.11.
1 Introduction
In multi-hop ad hoc network, throughput and fairness performances are very
important. However, IEEE 802.11 standard [1] does not provide good throughput
and fairness for stations in some asymmetry topologies [2, 3]. When stations
cooperate to forward packets from other stations through the ad hoc network,
each station has to transmit both the direct flow, which is generated by the
station, and forwarding flows, which are generated by its neighboring stations.
The station also shares the channel capacity with its neighboring stations. The
contentions in the MAC and link layers affect the performance of the network.
Due to the MAC layer contention, the allocated bandwidth for each station
cannot ensure per-flow fairness. In the link layer, the direct flow and forwarding
flows contend for the buffer space. Obviously, the direct flow gets more advantage
than forwarding flows [4–6].
In this paper, we propose a new scheme in order to achieve good fairness in
both the MAC and link layers. We also aim to ensure good throughput perfor-
mance and help disadvantaged flows get better throughput. A modification of
the MAC layer will let each station estimate network condition and determine
fair bandwidth allocation for them. By tuning back-off time, they can get suit-
able allocated bandwidth. In the link layer, we use Round Robin (RR) queue
for each flow, and monitor output packet to ensure per-flow fairness. We use the
Network Simulator (NS) [7] to evaluate our proposed scheme in some unfairness
situations relying on some asymmetry topologies.
The rest of the paper is organized as follows: Section 2 reviews some related
works. Section 3 describes our proposed scheme. Section 4 evaluates our proposed
scheme by comparing with original FIFO scheduling in IEEE 802.11 standard [1],
Shagdar’s method [5] and PCRQ scheduling [8]. Finally, Section 5 concludes the
paper.
2 Related work
The fairness performance at the MAC layer has been an active research field in
the past several years. The protocols MACA [9] and its extension MACAW [10]
use the four-way RTS/CTS/Data/ACK handshake signals to reduce collisions
caused by hidden terminals in the network. The protocol MACAW has been stan-
dardized in the IEEE 802.11 [1] as Distributed Coordination Function (DCF).
However, the RTS/CTS scheme in DCF does not solve all unfairness in band-
width in case of asymmetric links. Several schemes for improving the fairness
of MAC protocol have been proposed in the literature [11–13]. Moreover, Li
et al. [14] investigated Extended Inter-Frame Spacing (EIFS) problem, i.e., the
fixed EIFS value leads to unfair bandwidth allocation for each stations. They
proposed flexible EIFS values based on a measurement of the length of Sensing
Range (SR) frame. However, the length of SR frame cannot be always recognized
because of the spatial reuse of the bandwidth. Moreover, those research mainly
consider the MAC layer fairness problem, they do not consider per-flow fairness
problem.
Jangeun et al. [4] pointed out the weak point of FIFO scheduling in multi-hop
networks and then proposed various queuing schemes. Each scheme has offered
different degree of fairness. However, their research is based on the ideal MAC
layer fairness assumption that cannot be satisfied. Thus, their schemes do not
give good performance in the real networks. Shagdar et al. [5] and Izumikawa et
al. [6] also focused on the contention of direct and forwarding flows, and proposed
scheduling algorithms by using RR queue. The DCF mechanism are modified
in [5] to achieve the bandwidth utilization by sending all the packets at the head
of RR queues continuously without delay by back-off algorithm. However, their
solutions have the same problem as [4] due to the unsatisfactory assumption that
the MAC layer gives fair bandwidth allocation. PCRQ scheduling [8] also uses
RR queue with three algorithms to control input/output packets and the turn
of RR queue. It achieves good result in per-flow fairness. In PCRQ scheduling,
the MAC layer fairness is also improved indirectly because PCRQ scheduling
gives a delay for a packet from heavy flows in order to help disadvantaged flows
get a chance in using channel bandwidth. However, it slightly degrades the total
throughput performance.
3 Proposed Scheme
The reason of unfairness problem in multi-hop ad hoc network is due to both
the MAC and link layers contention. In this section, we propose a scheme to
improve per-flow fairness in wireless ad hoc networks. In the MAC layer, we
make a minor modification to the original DCF channel access mechanism to
help each station can achieve suitable allocated bandwidth. In the link layer, we
use RR queue with a monitoring algorithm to ensure fair between flows.
3.1 MAC Layer Modification
In IEEE 802.11, the DCF algorithm behaves unfair for throughput and allocated
bandwidth for asymmetric topology. To avoid this, our solution tries to obtain
distributed fair allocated bandwidth by controlling back-off time. By decreasing
back-off time, the channel access frequency of a station is increased and also the
given bandwidth for a station will be increased and vice versa.
MAC layer modification lets each station determine the current allocated
bandwidth and the fair allocated bandwidth for themselves. If a station finds
out its given bandwidth less than its fair allocated bandwidth, it will increase
its channel access frequency in order to achieve better allocated bandwidth. To
evaluate the current utilization bandwidth of a station, the station measures
time, which uses for transferring packets during an evaluating time.
Current utilization =T ransf erring time
Evaluating time (1)
To determine a fair allocated bandwidth for a station, the station observes
the channel and counts the number of flows in its carrier sensing range. The
flows in the transmission range are identified with MAC and IP addresses of
both source and destination. The flows, which are out of its transmission range
but in its carrier sensing range, are considered as one flow. The station also
examines the number of sending flows, which it is generating and forwarding.
The ratio of the number of sending flows to the number of total flows in its
carrier sensing range should be proportion of fair share bandwidth.
F air utiliz ation =Sending flows
T otal f lows (2)
When a station finds out that the current utilization in (1) is smaller than the
fair utilization in (2), it means that is not fair for the station in using bandwidth.
The station will try to increase the channel access frequency by turning its back-
off time as the following change of contention window size.
CWnew =C urrent utilization
F air utiliz ation CW (3)
3.2 Link Layer Modification
In the link layer, the direct flow’s queue tends to occupy completely the buffer
space and to seize almost bandwidth from forwarding flows. Therefore, the pro-
posed scheme monitors output packets from RR queues to ensure them using
bandwidth fairly. A heavy offered load flow is required sending some packets with
longer back-off time, and then more bandwidth is left for receiving packets from
its neighboring stations. Thus, throughputs of forwarding flows are improved. A
packet at the head of the queue for flow iis marked when it is dequeued, at the
following probability;
Pi marked =
0,if qleni≤ave
γqleni−ave
(n−1)ave ,if qleni> ave (4)
where γis an output weight constant. Packets from a heavy offered load flow may
be marked with probability in range 0 to γ. In case one queue is full while other
queue is empty, packets may be marked with probability γ. If the queue length
is smaller or equal to the average queue length, all packets will be dequeued
without being marked. When packet is marked, a cross-layer signal is sent to the
MAC layer to require send the packet with longer back-off time:
CWnew =κ∗C W (5)
where κis a delay weight constant.
4 Performance Evaluation
We now evaluate the performance of our proposed scheme by comparing with the
original FIFO scheduling in IEEE 801.11 standard [1], Shagdar’s method [5] and
PCRQ scheduling [8] on various topologies of multi-hop wireless ad hoc networks.
We use Network Simulator (NS-2) [7] for evaluation. The simulation parameters
are shown in Table 1. In PCRQ scheduling, we use the same parameters as in [8]:
the input weight constant α= 2.0, the hold weight constant β= 0.3, the output
weight constant γ= 0.3 and delay time δ= 1[ms]. In the proposed scheme, we
set the output weight constant γ= 0.3 and the delay weight constant κ= 2.
Table 1. Parameters in the simulation.
Channel data rate 2[Mbps]
Antenna type Omni direction
Radio Propagation Two-ray ground
Transmission range 250[m]
Carrier Sensing range 550[m]
MAC protocol IEEE 802.11b (RTS/CTS is enable)
Connection type UDP/CBR
Buffer size 100[packet]
Packet size 1[KB]
Simulation time 100[s]
The fairness and throughput performance metrics are evaluated.
–Fairness index: We use the fairness index, which is defined by R. Jain [15]
as follows:
Fairness Index = (Pn
i=1 xi)2
n·Pn
i=1 x2
i
(6)
where nis the number of flows, xiis the throughput of flow i. The result
ranges from 1/n to one. In the best case, i.e., throughput of all flows are
equal, the fairness index achieves one. In the worst case, the network is
totally unfair, i.e, one flow gets all capacity while other flows get nothing,
the fairness index is 1/n. In this paper, the fairness index is evaluated based
on goodput at the destination station.
–Total throughput: The average of total goodput of all flows in the network.
4.1 Scenario-1
Scenario-1 includes a chain of three stations with two flows. The coordinates
of stations are shown in Fig. 1. This topology is also known as Large-EIFS
problem [14], which is described in Fig. 2. In this scenario, stations S1 and S2 are
in one transmission range. Stations S1 and R are also in another transmission
range. Stations S2 and R are out of transmission range but in carrier-sensing
range. At the last state of four-way handshaking process from sender S1 to
receiver R, R sends an ACK frame in reply to a data frame from S1, then S2
detects the ACK frame, but cannot decode it. Thus, S2 must wait an EIFS
before accessing the channel, while S1 waits a DIFS, which is much smaller than
the EIFS. Li et al. [14] has proved that allocated bandwidth for S1 is four times
greater than S2 because of the Large-EIFS problem.
RS2 S1
Flow 1
Flow 2
Y-Axis
X-Axis
(200,0)(400,0) (0,0)
Fig. 1. Scenario-2: The Large-EIFS problem.
DIFS
ACK
SIFS CTS
SIFS DATA DIFS
DIFS Back-off
Back-off
Back-offRTS
SIFS
EIFS
NAV (RTS)
S 1
R
S 2
EIFS
Fig. 2. Unfairness in bandwidth due to Large-EIFS problem.
We examine network performances in this scenario by letting the stations
S1 and S2 generate traffic at the same offered load G to R. The performance
metrics are evaluated versus offered load G.
Fairness indices are shown in Fig. 3(a). When the offered load is small, all
scheduling methods get perfect fairness index. When the offered load becomes
larger, because a common queue is used in FIFO scheduling, the direct flow
gradually occupies completely the buffer space then the fairness index becomes
very bad. In Shagdar’s method [5], even different queue is used for each flow,
but the throughput of the forwarding flow is limited by the allocated bandwidth
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Offered load G [Mbps]
Fairness index
FIFO scheduling
Shagdar's method
PCRQ scheduling
Proposed scheme
(a) Fairness index
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Offered load G [Mbps]
Throughput [Mbps]
FIFO scheduling
Shagdar's method
PCRQ scheduling
Proposed scheme
(b) Total throughput
Fig. 3. Simulation results in Scenario-1
of S2 which is too much smaller due to Large-EIFS problem [14]. Thus, the
fairness index in Shagdar’s method is not good. In PCRQ scheduling [8], input
and output packets to RR queues is monitored, the per-flow throughput becomes
fairer and also the allocated bandwidth at the MAC layer is improved indirectly.
Thus, PCRQ scheduling achieves good fairness index. In the proposed scheme,
both the MAC and link layers fairness are considered, we also achieve very good
fairness index.
The total throughputs of all flows are shown in Fig. 3(b). When the offered
load is small, throughputs in all method are similar. When the offered load
becomes greater, PCRQ scheduling uses bandwidth slightly less efficiently than
the others do. Because PCRQ scheduling makes a delay for sending to give
chance for receiving in order to achieve the MAC layer fairness indirectly. In the
proposed scheme, S2 increase its channel access frequency to get the MAC layer
fairness. It also improves network utilization and the proposed scheme achieves
better throughput than PCRQ scheduling.
4.2 Scenario-2
Scenario-2 is three-pairs scenario. The coordinates of stations are shown in Fig. 5.
In this scenario, stations S1-S2 and S2-S3 are out of transmission range but in
carrier sensing range. Stations S1-S3 are out of carrier sensing range and the
two external pairs S1-R1 and S3-R3 are independent from each other. Thus, two
external pairs contend bandwidth with only the central pair S2-R2 while the
central pair contends with both external pairs. In this topology, the central pair
cannot access the medium in saturated state.
We examine fairness problem in this topology by letting the stations S1, S2
and S3 generate traffic at the same offered load G to R1, R2 and R3, respectively.
The performance metrics are evaluated versus offered load G.
Fairness indices are shown in Fig. 5(a). When the offered load is small, all
scheduling methods get perfect fairness index. When the offered load becomes
larger, FIFO scheduling does not help the central pair to access the medium.
However, both PCRQ scheduling and Shagdar’s method also do not have bet-
ter results than FIFO scheduling. The reasons are explained as follows. PCRQ
scheduling only works in the link layer, it does not have information of flows
out of transmission range. Therefore, it cannot improve the MAC layer fairness.
Shagdar’s method works in both the MAC and link layers. It modifies back-off
algorithm in order to achieve per-flow fairness and network utilization. However,
Shagdar’s method cannot ensure the MAC layer fairness and it fails to achieve
S1
R3
S2
R2
S3
Flow 1
Flow 2
Flow 3
Y-Axis
X-Axis
(600,200) (0,200)
(0,0)(600,0)
(300,200)
R1
(300,0)
Fig. 4. Scenario-2: three pairs scenario.
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Offered load G [Mbps]
Fairness index
FIFO scheduling
Shagdar's method
PCRQ scheduling
Proposed scheme
(a) Fairness index
0
0.5
1
1.5
2
2.5
3
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Offered load [Mbps]
Throughput [Mbps]
FIFO scheduling
Shagdar's method
PCRQ scheduling
Proposed scheme
(b) Total throughput
Fig. 5. Simulation results in Scenario-2
per-flow fairness in this topology. In the proposed scheme, the central pair finds
out that it gets less bandwidth than its fair allocated bandwidth. It will improve
the channel access frequency by reducing its back-off time. Thus, the proposed
scheme can achieve the good MAC layer fairness and per-flow fairness.
The total throughputs of all flows are shown in Fig. 5(b). When the offered
load is small, throughputs in all method are similar. When the offered load
becomes greater, the proposed scheme achieves total throughput smaller than
the other methods as the trade-off between fairness and throughput. In the
other methods, the central pair cannot access the channel bandwidth and two
external pairs can use maximum channel bandwidth. The total throughput can
be twice of maximum channel bandwidth. In the proposed scheme, the MAC
layer fairness is ensured, the central pair can achieve a half of maximum channel
bandwidth. It leads two external pairs also can use a half of maximum channel
bandwidth. Thus, the total throughput is only one and a half of maximum
channel bandwidth. It is obviously smaller than in the other methods.
4.3 Scenario-3
Scenario-3 includes a chain of five stations with four flows. The coordinates of
stations are shown in Fig. 6. The stations S1, S2, S3, and S4 generate traffic at
the same offered load G to R.
The performance results are shown in Fig. 7(a) and 7(b). The results show
that both PCRQ scheduling and the proposed scheme get better fairness perfor-
mance than the others. The proposed scheme also have advantage in throughput
performance by comparing with PCRQ scheduling.
RS2 S1S4 S3
Flow 1
Flow 2
Flow 3
Flow 4
Y-Axis
X-Axis
(200,0)(400,0) (0,0)(600,0)(800,0)
Fig. 6. Scenario-3: a five-station chain with four flows.
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Offered load G [Mbps]
Fairness index
FIFO scheduling
Shagdar's method
PCRQ scheduling
Proposed scheme
(a) Fairness index
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Offered load [Mbps]
Throughput [Mbps]
FIFO scheduling
Shagdar's method
PCRQ scheduling
Proposed scheme
(b) Total throughput
Fig. 7. Simulation results in Scenario-3
4.4 Scenario-4
Scenario-4 is a grid scenario with high station density and high traffic density as
in Fig. 8. In this scenario, the columns are separated by distance greater than
the transmission range but smaller than the carrier sensing range. The stations
in a column generate traffic at the same offered load G to the receiver in the
same column. This topology faces both Large-EIFS and Three-pairs unfairness
problems at the MAC layer.
S1
R3
S3
R2
S5
Y-Axis
X-Axis
(600,200)
R1
S2S4S6
(0,0) (0,200) (0,400)
(300,0)
(600,0)
(300,200) (300,400)
(600,400)
Fig. 8. Scenario-4: Grid scenario.
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Offered load G [Mbps]
Fairness index
FIFO scheduling Shagdar's method
PCRQ scheduling Proposed scheme
(a) Fairness index
0
0.5
1
1.5
2
2.5
3
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Offered load [Mbps]
Throughput [Mbps]
FIFO scheduling
Shagdar's method
PCRQ scheduling
Proposed scheme
(b) Total throughput
Fig. 9. Simulation results in Scenario-4
The fairness between six flows are shown in Fig. 9(a). At the high offered
load, each method achieves different fairness level due to contention between the
forwarding flow and direct flow at the link layer, Large-EIFS and Three-pairs
problems at the MAC layer. The fairness index of FIFO scheduling is the worst
because of both unfairness problems in the MAC and link layers. Shargdar’s
method achieves fairness better than FIFO scheduling due to using RR queue.
However, it still faces the MAC layer unfairness problem. PCRQ scheduling can
solve the unfairness in the link layer and Large-EIFS problems at the MAC layer.
Our proposed method can improve fairness both at the MAC and link layers.
Therefore, we achieve the best fairness in all methods.
The total throughputs of all flows are shown in Fig. 9(b). At the high offered
load, the proposed scheme achieves total throughput is smaller than the other
methods as the same reason with Scenario-2.
5 Conclusion
Our proposed scheme is divided in two parts. One works on the MAC layer. We
slightly modified back-off algorithm in order to achieve fair allocated bandwidth
for each station. The other works on the link layer. We controlled RR queue in
order to help each flow use bandwidth fairly. The cross-layer signal is sent from
the link layer to the MAC layer to required the MAC layer sends the packet
from the heavy offered load flow with longer back off time. The parameter γcan
be chosen any value from zero to one and the parameter κcan be chosen any
value greater than one. The same reason with PCRQ scheduling, the large value
of these parameters may degrade bandwidth utilization. In our simulations, we
chose γequal to 0.3 and κequal to 2.
In the simulation results, we evaluate our method by comparing with some
other methods. The results showed that our method achieved good performances
in Large-EIFS problem, three-pairs problem scenarios and also complex topolo-
gies as long-station chain and grid scenarios. In these topologies, FIFO schedul-
ing has only one queue, and then it cannot solve the link layer contention. Shag-
dar’s method uses RR queue, but also works ineffectively because the allocated
bandwidths at the MAC layer are not suitable for forwarding and direct flows
in the link layer. PCRQ scheduling can achieve good per-flow fairness. However,
it fails to keep good per-flow fairness of flows, which do not come to the link
layer. The proposed scheme works in both the MAC and link layers for aiming to
improve both the MAC and link layer fairness. Therefore, we have good results
in per-flow fairness even in the MAC layer unfairness topologies. Moreover, the
proposed scheme gets better throughput performance than PCRQ scheduling
compared to link layer topologies as Large-EIFS and long-station chain topolo-
gies. Because PCRQ scheduling tries to get the MAC layer fairness indirectly by
giving a delay of heavy offered load flow to give a chance for forwarding flows.
It affects to throughput performance. The proposed scheme adjusts back-off
time to let disadvantaged station can get better channel bandwidth. Therefore,
throughput performance in the proposed scheme also improves. In the MAC layer
unfairness topologies, fairness between stations is improved by our method then
total throughput is reduced as the trade-off between fairness and throughput.
References
1. Department, I.S.: IEEE 802.11 wireless lan medium access control (MAC) and
physical layer (PHY) specifications. ANSI/IEEE Standard 802.11 (1999)
2. Xu, S., Saadawi, T.: Does the ieee 802.11 mac protocol work well in multihop
wireless ad hoc networks? Communications Magazine, IEEE 39(6) (2001) 130–137
3. He, J., Pung, H.K.: One/zero fairness problem of MAC protocols in multi-hop
ad hoc networks and its solution. Proceedings of the International Conference on
Wireless Networks (2003) 479–485
4. Jangeun, J., Sichitiu, M.: Fairness and qos in multihop wireless networks. IEEE
Vehicular Technology Conference 5(2003) 2936–2940
5. Shagdar, O., Nakagawa, K., Zhang, B.: Achieving per-flow fairness in wireless ad
hoc networks. Elec. Comm. in Japan, Part 1 89(8) (2006) 37–49
6. Izumikawa, H., Sugiyama, K., Matsumoto, S.: Scheduling algorithm for fairness
improvement among subscribers in multihop wireless networks. Elec. Comm. in
Japan, Part 1 90(4) (2007) 11–22
7. : The Network Simulator: ns-2. http://www.isi.edu/nsnam/ns/
8. Giang, P.T., Nakagawa, K.: Scheduling algorithms for performance and fairness
over 802.11 multi-hop wireless ad hoc networks. In: IEICE Tech. Rep. NS2008-21
(June 2008) 19–24
9. Karn, P.: Maca: A new channel access method for packet radio. In: ARRL/CRRL
Amateur Radio 9th computer Networking Conference. (1990) 134–140
10. Bharghavan, V., Demers, A., Shenker, S., Zhang, L.: MACAW: a media access
protocol for wireless lan’s. In: Proceedings of the conference on Communications
architectures, protocols and applications, ACM (1994) 212–225
11. Nandagopal, T., Kim, T.E., Gao, X., Bharghavan, V.: Achieving mac layer fair-
ness in wireless packet networks. In: Proceedings of the 6th annual international
conference on Mobile computing and networking, ACM (2000) 87–98
12. Bensaou, B., Wang, Y., Ko, C.C.: Fair medium access in 802.11 based wireless
ad-hoc networks. In: Proceedings of the 1st ACM international symposium on
Mobile ad hoc networking & computing, IEEE Press (2000) 99–106
13. Jiang, L.B., Liew, S.C.: Improving throughput and fairness by reducing exposed
and hidden nodes in 802.11 networks. IEEE Transactions on Mobile Computing
7(1) (2008) 34–49
14. Li, Z., Nandi, S., Gupta, A.K.: Ecs: An enhanced carrier sensing mechanism for
wireless ad-hoc networks. Computer Communication 28(17) (2005) 1970–1984
15. R.Jain, Chiu, D.M., Hawe, W.: A quantitative measure of fairness and discrimina-
tion for resource allocation in shared conputer systems. Technical Report TR-301,
DEC Research Report (1984)
