A novel Service Level Agreement based algorithm for ... - CiteSeerX

A novel Service Level Agreement based algorithm for Differentiated Services enabled Ethernet PONs Dawid Nowak, Philip Perry

John Murphy

Performance Engineering Laboratory School of Electronic Engineering Dublin City University Ireland Email: {nowakd, perryp}@eeng.dcu.ie

Performance Engineering Laboratory Department of Computer Science University College Dublin Ireland Email: [email protected]

Abstract— We present a Dynamic Bandwidth Allocation algorithm that supports Service Level Agreements in Ethernet Passive Optical Networks and show that it provides a reliable protection of agreed traffic parameters. We include results of simulation experiments.

1. I NTRODUCTION An optimal bandwidth assignment in Ethernet Passive Optical Networks (EPONs) is not a straightforward goal to achieve. A good algorithm has to provide best utilization of available bandwidth as well as it should support various classes of traffic. The latter is especially hard to achieve as voice and data services demand a completely different service. Various schemes have been proposed in recent years in literature. In [1], a Dynamic Bandwidth Allocation (DBA) algorithm called ”Interleaved Polling with Adaptive Cycle Time (IPACT)” was presented. DBA combined with priority scheduling algorithms was studied in [1], [2]. The performance of a DBA with cyclic polling and some support for Differentiated Services (DiffServ) [3] was studied in [4]. Here, we present a novel approach that tries to find an answer to following problems: • An Optical Network Unit (ONU) should be simple and inexpensive. The access control or queue management should be a task of an Optical Line Terminator (OLT) which is usually placed in a local exchange. This allows for more generic, hence cheaper ONUs. • A bandwidth assignment must provide protection of traffic parameters agreed in a Service Level Agreement (SLA) between a customer and the network provider. This is extremely important as EPON, with the optimal Quality of Service (QoS), must support various types of traffic often requiring different channel properties. • Possible updates and modifications to the SLA should not interrupt a normal network operation. 2. SLA AWARE DYNAMIC BANDWIDTH A LLOCATION A LGORITHM (SLA-DBA) Here, we present the algorithm that is based on a proportional allocation of bandwidth per traffic class as

that provides the best utilization of available resources. To fully support SLAs, bandwidth constraints agreed in a service contract are used to ensure that the just amount of bandwidth is assigned to a particular class. Let Qi (j) be the amount of the bandwidth requested for queue j by ONU i and βki (j) is the bandwidth allocated i to this queue in step k of the algorithm. Let γmin (j) and i γmax (j) be the minimum and maximum of the bandwidth guaranteed to the particular queue. In phase I, the OLT assigns bandwidth proportionally to reported queue length Qi (j). βIi (j) = P

1 · Qi (j) i i,j Q (j)

(1)

In the second phase the constraints given in the SLA are applied. Three distinct situations have to be considered: i 1) βIi (j) ≥ γmax (j) – Assigned bandwidth has exceeded the amount promised in the SLA. The bandwidth allocated to a particular queue is thus i i reduced to βII (j) = γmax (j). i i i 2) βI (j) ≥ γmin (j) and βIi (j) < γmax (j) – Requested bandwidth is within the limits of the SLA. i No changes are made and βII (j) = βIi (j) i i 3) βI (j) < γmin (j) – In a situation where Qi (j) > i β i (j) bandwidth assigned is equal to βII (j) = i Q (j) as it is smaller than agreed in the SLA. i Otherwise no changes are made and βII (j) = βIi (j)

Bandwidth that is not allocated during the second step is shared among all queues in phase three. The amount of bandwidth assigned to a queue could be thus expressed as: βex i i (2) βIII (j) = βII (j) + i βI (j) 3. P ERFORMANCE E VALUATION To measure the performance of each bandwidth allocation algorithm we designed an event-driven C++ based EPON simulator. We used 16 ONUs connected in a tree topology to a single OLT operating at a speed of 1Gb/s. Each ONU had three queues with an independent buffering space. The guard time between transmissions from different ONUs was set to 1µs and the value of

TABLE I

0.01

SLA-DBA DBA SBA

0.009

AVERAGE DELAY FOR EF CLASS FOR ONU S IN (I) AND OUTSIDE (II) OF THE LIMITS OF THE SLA. T IMES IN m S .

Average delay (s)

0.008 0.007

(a) EF class Bandwidth utilization % 40 60 80 100 115 130 145

0.006 0.005 0.004 0.003 0.002

SLA (I) 2.0686 2.0853 2.1027 8.9956 11.037 11.835 12.057

0.001 0

10

20

30

40

50

60

70

80

Bandwidth utilization % 40 60 80 100 115 130 145

(a) EF 0.01

SLA-DBA DBA SBA

0.009 Average delay (s)

0.008

DBA (I) (II) 2.1131 1.9735 2.1618 2.0292 2.2965 2.1017 33.730 31.359 146.77 135.63 206.13 192.40 249.29 233.77

SBA (I) (II) 1.9876 1.9333 7.8035 7.8171 9.2451 9.6888 9.3852 14.486 11.183 15.827 9.5653 26.923 9.6108 111.95

(b) AF class

90 100

Load

(II) 1.9416 1.9527 1.9834 9.6421 19.752 24.548 105.566

SLA (I) 3.5930 3.5982 3.6292 14.886 19.217 18.807 19.371

(II) 3.3114 3.4472 3.5431 23.787 66.639 90.011 106.96

DBA (I) (II) 3.6536 3.3802 3.7416 3.5348 3.8126 3.6923 41.303 42.131 153.51 149.86 207.36 208.20 261.91 245.01

SBA (I) (II) 1.9457 5.1191 7.9578 72.877 11.565 234.71 28.487 293.57 141.98 259.78 107.39 355.90 180.95 373.77

0.007 0.006 0.005 0.004 0.003 0.002 0.001 0

10

20

30

40

50

60

70

80

90 100

Load

(b) AF Fig. 1. Algorithms performance comparison. Average delays for EF (a) and AF (b).

Inter-Frame Gap (IFG) between Ethernet packets was 96 bits. It was shown that most network traffic (i.e., http, ftp and VBR services) was best characterized by self-similarity and long-range dependence [5]. To model a high priority Expedited Forwarding (EF) class of traffic (e.g., voice applications) a Poisson distribution is generally used. In our simulation EF traffic constituted 20% of the bandwidth. The remaining bandwidth was divided equally between Assured Forwarding (AF) and Best Effort (BE) classes. The length of high priority traffic packets was fixed at 70 bytes. The length of AF and BE packets was uniformly distributed between 64 and 1518 bytes. The granting cycle time was variable depending on the amount of bandwidth requested in the previous granting cycle. In any case the granting cycle could not be smaller than 2ms or larger than 20ms. We compared the performance of the SLA-DBA algorithm against the performance of SBA (static bandwidth assignment) and DBA algorithms. In Fig. 1 we show average delays for EF (a) and AF (b). It is clear that the SLA-DBA scheme gives better results then SBA and DBA algorithms. To compare the behavior of algorithms where some classes exceed the limits imposed in the SLA, we increased transmission rate of 8 ONUs by 50%. The results of the simulations are presented in Table I.

The SBA algorithm showed the best performance, although this comes at the price of worse bandwidth utilization. The DBA algorithm provided no protection as the difference in average delay time between faster and slower ONUs was negligible. The SLA-DBA algorithm provided good protection and classes were always allocated the amount of bandwidth which was agreed in the SLA. For small loads it was possible that more bandwidth was allocated to classes that exceeded their SLAs. This is due to a proportional bandwidth allocation in the first phase. Despite this, classes that conform to their SLAs, receive enough bandwidth. 4. C ONCLUSION We presented a novel DBA algorithm with a full support for SLAs. We included the results of extensive simulation experiments that show that SLA-DBA offers a good performance in terms of average packet delay. On the contrary to other popular algorithms, the SLA-DBA scheme provides a reliable protection mechanism of the service contract agreed in the SLA. This unique feature makes it especially suitable for DiffServ enabled EPONs. R EFERENCES [1] G.Kramer, B.Mukherjee, S.Dixit, Y.Ye, and R.Hirth, “On supporting differentiated classes of service in EPON-based access network,” OSA Journal of Optical Networking, vol. 1, no. 8/9, pp. 280–298, 2002. [2] Ch.M.Assi, Y.Ye, S.Dixit, and M.A.Ali, “Dynamic Bandwidth Allocation for Quality-of-Service Over Ethernet PONs,” IEEE Journal on Selected Areas in Communications, vol. 21, no. 9, pp. 1467– 1477, Nov. 2003. [3] S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, and W.Weiss, “RFC-2475 - An Architecture for Differentiated Services,” 1998, http://www.ietf.org/rfc/rfc2475.txt. [4] Su-il Choi, “Cyclic Polling-Based Dynamic Bandwidth Allocation for Differentiated Classes of Service in Ethernet Passive Optical Networks,” Photonic Network Communications, vol. 7, no. 1, pp. 87–96, 2004. [5] W. Leland, M. Taqqu, W. Willinger, and D. Wilson, “On the SelfSimilar Nature of Ethernet Traffic (extended version),” IEEE/ACM Transactions on Networking, vol. 2, no. 1, pp. 1–15, Feb. 1994.