вход по аккаунту



код для вставкиСкачать
OFC 2014 © OSA 2014
Integrated Packet/Circuit Hybrid Network Field-Trial
Demonstrating Sub-Wavelength Aggregation
Steinar Bjornstad1, 2, Raimena Veisllari1, Jan P. Braute2 and Kurosh Bozorgebrahimi3
Department of Telematics, Norwegian University of Science and Technology, O.S. Bragstads 2b, Trondheim, Norway
TransPacket, Drammensveien 126, 0277 Oslo, Norway
UNINETT, Abels gate 5, 7465, Trondheim Norway
Abstract: We report aggregation of sub-wavelengths in an integrated packet/circuit hybrid optical
network. Aggregation of packet streams with circuit quality of service combined with statistical
multiplexing enables packet delay variation of only 15ns at 82.4% throughput.
OCIS codes: (060.4253) Networks, circuit-switched; (060.4259) Networks, packet-switched
1. Introduction
With the ever increasing channel bitrates in optical communication systems now targeting Tb/s capacities, the need
for splitting the bandwidth into sub-wavelength capacities with the characteristics of wavelengths for aggregation
purposes becomes increasingly important. Optical Transport Network, G.709 (OTN) offers a circuit switched
approach to aggregation and switching of sub-wavelengths allowing zero packet loss, low-latency, ultra-low packetdelay variation and synchronization transport. These characteristics enable support for a variety of applications and
systems needing timing-critical transport at the underlying transport layer. Strict timing is a key requirement for e.g.
mobile backhaul networks, as well as for private lines in sub-wavelengths, allowing full isolation also on timing
[1][2]. Packet networks on the other hand, offer a higher throughput by using statistical multiplexing, but do not
match OTN network timing-performance [3]. Thus, there is a need for a single integrated network that combines the
properties of the circuit switched OTN with the packet switched network into a single network enabling both: (1)
circuit quality transport of the demanding services and (2) high throughput efficiency of packet networks.
Integrated hybrid optical networks (IHON) [4], also known as Fusion, merge the circuit and packet network in
the same wavelength in a time-interleaved manner without using fixed time-slots. Earlier IHON experiments have
demonstrated how the timing is preserved on a single Guaranteed Service Traffic (GST) packet stream while adding
statistically multiplexed (SM) packet-switched traffic [5]. In this paper we show for the first time how several lower
bitrate GST packet streams, i.e. sub-wavelengths, can be aggregated into, and de-aggregated from, a higher bitrate
GST packet stream following a wavelength [6]. We demonstrate experimentally that GST enables sub-wavelengths
with circuit-switched quality of service (QoS), i.e., absolute transfer guarantees with no packet loss, while packet
delay and packet delay variation (PDV) have the characteristics of circuit switching. Thus, IHON offers transport for
the most demanding services like real-time traffic, synchronization, private-lines and control information. For
maintaining the circuit QoS, i.e. avoiding PDV, GST preserves both packet-lengths and packet gaps when being
aggregated into a higher bitrate stream. This enables accurate reconstruction of the stream at the aggregation end.
Furthermore, any capacity not utilized by the GST packets is identified in the packet stream as idle time-gaps. The
lightpath utilization is increased by filling these gaps with packets from a statistical multiplexed (SM) class with best
effort QoS.
2. IHON node and aggregation
Figure 1(a) shows a block diagram of the IHON node from TransPacket, applied in the experiment. Five GST 1 Gb/s
Ethernet input streams are aggregated into one 10 Gb/s Ethernet output link. Each of the inputs is logically divided
into a container corresponding to the duration of three 2048 Byte maximum transmission unit (MTU). These five
bursts are then aggregated into the 10 Gb/s channel frame. The nodes use VLAN tags to separate the traffic streams,
GST and SM. Hence, the GST packets arriving into a 1 Gb/s input are tagged, their inter-packet gap in bytes is
extracted and the packets are aggregated into their corresponding container maintaining the gaps between the
packets. Each sub-wavelength stream follows a dedicated lightpath in the network with a light processing in the
nodes, easing processing in any intermediate nodes. Streams are identified by processing only the VLAN tag of the
packets in the streams. At the destination, the packets are extracted from their container and the packet gaps are used
to precisely reconstruct the stream.
At each node, before accessing the output channel, the aggregated GST traffic passes through a fixed delay δ of
7.68 µs. It corresponds to the transmission time of an SM MTU of 9600 Byte and prevents preemption of SM
978-1-55752-993-0/14/$31.00 ©2014 Optical Society of America
OFC 2014 © OSA 2014
packets by incoming GST packets, e.g. SM packets are not scheduled when a GST packet entering the delay is
detected. The effect of δ on the end-to-end delay is deterministic, and it allows the GST gap detector to detect the
duration of idle time-gaps between GST packets in the 10 Gb/s channel. This information is used by the SM
scheduler checking the SM input queue(s) for a packet of suitable size fitting the gap. If a suitable packet is found,
the SM packet is inserted without affecting the timing of the packets in the GST stream, while increasing the
throughput in the wavelength.
Fig.1: (a) Block diagram of the IHON node with five GST streams (1 Gb/s Ethernet) aggregation and SM insertion. (b) The field-trial setup with
two prototype H1 nodes from TransPacket and Spirent SPT-2000 packet generator/tester;
3. IHON Test-bed Setup
The main goal of the experiment is to demonstrate that IHON provides sub-wavelength transport with circuit QoS
through the GST aggregation and de-aggregation: Bounded low deterministic delay, ultra-low packet delay variation
and no packet loss. In addition, previously demonstrated IHON characteristics should be maintained, i.e. the GST
circuit transport is independent of the insertion of the statistically multiplexed traffic that is applied for increasing
the lightpath utilization efficiency [5]. In the carrier provider network of UNINETT, two IHON nodes were
connected through a fiber link of 3.25 km with a 10 Gb/s Ethernet wavelength, as illustrated in Fig. 1(b). Five GST
streams were added through the packet generator to the 1 Gb/s ports of 1 , aggregated into the 10 Gb/s wavelength
channel, de-aggregated by 2 to their corresponding ports and received for measurement back to the tester. The
Ethernet packet length follows a tri-modal empirical distribution taken from Internet measurements. The offered
load  of each GST stream  is varied by changing the inter-packet length providing a total GST offered load in the
channel  = ∑  . One SM stream was added from the tester to the second 10 Gb/s port of 1 transmitted on the
free time-gaps in the transport wavelength between the nodes, and sent back to the tester from 2 . The SM offered
load  was varied to demonstrate the increase in the achievable maximum throughput without any impact on the
GST streams.
4. Results and Discussion
The performance results of the GST aggregation scheme were gathered both through reference tests in the lab and in
the field-trial depicted in Fig. 1(b). All the GST streams had an equal average load  normalized over the 10 Gb/s
wavelength, offering a total GST load  = 5 in the channel.  was varied from 0.01 to 0.5 with and without
inserting SM traffic and both sets of results were consistent: the measured GST delay was found constant,
independently of the GST or SM load in the system. The average GST delay of the end-to-end streams is 67.22 µs
where 16.22 µs correspond to the propagation delay in the fiber link and 51 µs to the nodal delays. From the latter,
7.68 µs corresponds to  and 43.3 µs to the nodes processing delay. The average packet delay variation, i.e. delay
variation between consecutive packets in a stream, is 15 ns with a measured maximum of 160 ns. The minimum and
maximum (boundary) delay values registered over all tests and loads differ with an average of 320 ns, corresponding
to the peak PDV for GST.
The GST and SM delay results are depicted in Fig. 2. The total offered load  in the system is equally offered
by the GST aggregate and SM, so that  =  +  and  =  . The boundary SM delay values can be seen
plotted together with the average and illustrate how the increase in GST load influences the SM performance, i.e. the
average delay and bounds rise. We observe that as both the GST and SM load increase, the GST average delay is
OFC 2014 © OSA 2014
kept constant. When  reaches 0.746 with 0.375 GST load, the system goes into saturation. The SM traffic starts
experiencing congestion and we observe packet losses, but the GST characteristics/QoS parameters do not change.
We continue increasing the GST load at full capacity for each GST stream, i.e.  =0.5, without any impact on its
performance, while the SM packet delay and loss continues to rise. The sub-wavelength granularity transport has
absolute priority over SM and neither its delay, nor PDV nor loss are affected by the SM insertion. Thus, the
experimental results confirm that the GST streams maintain their characteristics, i.e. inter-packet length, during the
aggregation and transport through the IHON network. Furthermore, results show the isolation of the GST streams
from each other, i.e. no packet losses and the average delay remains constant regardless of the link condition.
TABLE I. Maximum carried SM in the wavelength [%]
Fig.2: Packet delay as a function of the normalized offered load on the 10
Gb/s Ethernet wavelength.
Number of GST streams: g
The maximum carried SM traffic in the system is given in Table I. The offered GST load  is varied from 10 %
to 50 % of the wavelength capacity and different combinations of the number  of GST streams offering this total
load are measured. There is a trade-off between the number of GST streams and the maximum achievable
throughput: A high number of GST streams for the same offered load  increases the fragmentation of the available
bandwidth. We observe that the SM insertion increases the wavelength utilization with a maximum of 32.4% for 0.5
GST load, up to 8.24 Gb/s total throughput. Hence, the network shows to perform as a statistical multiplexing packet
network with high utilization while still providing a service with circuit QoS properties.
5. Conclusions
We have demonstrated the first Ethernet circuit transport at sub-wavelength granularity on a wavelength through an
integrated hybrid circuit/packet network field-trial. Results confirm that IHON TransPacket prototype nodes
aggregate, transport and de-aggregate GST streams while: (1) maintaining the stream characteristic, i.e. its interpacket timing before aggregation, with ultra-low packet delay variation; (2) enabling circuit-switching properties
with a low deterministic delay and no packet loss. High throughput efficiency was demonstrated by adding packetswitched statistical multiplexed traffic on the common circuit/packet lightpath. A total channel utilization of 82.4 %
was reached without packet loss on the SM class. For all loads, no timing-impact on the GST sub-wavelengths was
found. The field-trial demonstrates the maturity of the IHON technology and its capability of removing the timing
obstacles for packet networks, matching the timing of OTN and legacy SDH/SONET networks. By applying IHON
networks, packet based networks are brought closer to fully replace the TDM based networks.
5. References
[1] Z. Ghebretensae, J. Harmatos, K. Gustafsson, "Mobile broadband backhaul network migration from TDM to carrier ethernet," IEEE
Communication Magazine, Vol 48(10), pp.102-109 (2010).
[2] S. Thiagarajan, M. Gemelos, M. Ma, "Transport Network Evolution for Advanced Services," OFC/NFOEC 2013, paper NTh4J.4.
[3] A. Gumaste, N. Krishnaswamy, “Proliferation of the optical transport network: A use case based study,” IEEE Communication Magazine,
Vol 48(9), pp. 54–61 (2010).
[4] C.M. Gauger, et. al, “Hybrid optical network architectures: bringing packets and circuits together,” IEEE Communication Magazine, Vol
44(8), pp. 36-42 (2006).
[5] S. Bjornstad, R. Veisllari, K. Bozorgebrahimi, “Long-haul Ethernet-Transport Integrated Packet/Circuit Hybrid Network Field-Trial,”
OFC/NFOEC 2013, paper PDP5A.8.
[6] S. Bjornstad, J.P. Braute, “Optical Switching”, Patent Application WO2012104623A3 (2013).
Без категории
Размер файла
430 Кб
ofc, w4h, 2014
Пожаловаться на содержимое документа