fs-PGBR: A Scalable and Delay Sensitive Cloud Routing Protocol Julien Mineraud, William Donnelly Sasitharan Balasubramaniam
Jussi Kangasharju Department of Computer Science University of Helsinki, Finland
Telecommunications Software & Systems Group Waterford Institute of Technology, Ireland
[email protected]
{jmineraud, wdonnelly, sasib}@tssg.org ABSTRACT This paper proposes an improved version of a fully distributed routing protocol, that is applicable for cloud computing infrastructure. Simulation results shows the protocol is ideal for discovering cloud services in a scalable manner with minimum latency.
Categories and Subject Descriptors C.2.2 [Network Protocols]: Routing Protocols
General Terms Performance
Figure 1: Cloud Computing Infrastructure
Keywords
2.
Scalable route discovery, Cloud computing infrastructure
1.
PROPOSED SOLUTION
2.1
MOTIVATION
PGBR
The original PGBR algorithm is a bio-inspired algorithm that is inspired from the bacteria motility process known as chemotaxis. The process of chemotaxis, enables bacteria to mobilize itself towards a destination by sniffing through an attractant gradient field formed in the environment. Mimicking this process, the PGBR algorithm generates an increasing chemical gradient field that attracts packets to a destination node. The equation for generating the gradient field can be represented as:
Cloud computing is recognized as the next generation infrastructure that would revolutionize the way users access services [3]. The driving force behind the concept of cloud computing has been fueled by the growth of computing devices (in particular mobile devices) as well as rich services which execute on power and memory constrained devices. The vision of cloud computing will enable users to offload these services into high powered storage facilities, where users will be able to access infrastructures, platforms, as well as software within the cloud. In essence, the cloud will represent a network of high powered data centers [2] that provides a multitude of services to end users based on a pay as you go model (Figure 1). Although the cloud will enable users to elastically consume resources, this will bring along new challenges for the underlying infrastructure. One of these challenges include a dynamic traffic pattern in the underlying network, which will require more scalable, efficient, and adaptive routing algorithms. In this paper, we propose a scalable routing algorithm called fast search - Parameterised Gradient Based Routing (fs-PGBR) that is fully distributed, scalable and reduces considerable latency during route discovery. The algorithm is an extension to the Parameterised Gradient Based Routing (PGBR) [1], which suffers from scalability performance once the network topology size increases.
Gdi (j) = αnl(j) + βlli (j) + γhdi (j)
(1)
where Gdi (j) is the gradient value of link i → j for a packet to destination d, nl(j) represents the node load value of neighbour node j, lli (j) is the load value of the link i → j, hdi (j) is the normalised hop count neighbour node j of i for destination d, and α, β and γ are weighting parameters. The weighting parameters provides an extra feature of service oriented routing, where different weight sets can correspond to different service types. The value for hdi (j) is calculated during the topology formation, and will remain static. The equation for hdi (j) is represented as: hdi (j) = 1 −
wd (j) , where W d = max(wd (k)), ∀k Wd
(2)
where wd (j) is the minimum distance of node j to destination d. While PGBR provides numerous benefits that are suitable for the underlying network supporting the cloud infras-
Copyright is held by the author/owner(s). SIGCOMM’12, August 13–17, 2012, Helsinki, Finland. ACM 978-1-4503-1419-0/12/08.
301
500 400 300 200
4
1 hop 2 hops 3 hops 4 hops 5 hops 6 hops 7 hops
250 Hops to destination
Hops to destination
300
1 hop 2 hops 3 hops 4 hops 5 hops 6 hops 7 hops
600
200 150 100 50
100 0 0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Gamma value (a) No. of hops to destination (PGBR)
3 2.5 2 1.5
0 0.1
PGBR fs-PGBR
3.5 Path length ratio
700
1 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.1
0.2
Gamma value (b) No. of hops to destination (fs-PGBR)
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Gamma value (c) Path length ratio
Figure 2: Performance evaluation of the new metric tructure, there is a slight shortfall. This shortfall is in the approach used to calculate hdi (j), which is not scalable once the topology reaches a large number of nodes. In particular, this could be an issue for large-scale cloud infrastructure supported by a network topology with a high number of nodes.
2.2
compared to the original approach, in particular for long distance sources. This is due to a number of factors, including minimal discovery rejection that may result from Time-To-Live (TTL) expiration. Consequently, the new extension also enables more diversity in the choice of parameters for fs-PGBR discoveries. Secondly, as shown in Figure 2 (c), the resulting path is maintained under smaller and more reasonable path length ratio compared to the original approach. This demonstrates the effectiveness of fs-PGBR and its suitability towards timely route discovery for cloud computing infrastructures.
Hop count value for fs-PGBR
In order to counter the scalability problem, we have modified the hdi (j) calculation, and this is represented as: hdi (j) =
max(wid (k)) − wid (j) , ∀k neighbour of i (3) max(wid (k)) − min(wid (k))
4.
CLOUD COMPUTING INTEGRATION
This modification enables each node i to compare the shortest path weight of its neighbour k and evaluate it against the maximum and minimum weights of its neighbours. In the case of max(wid (k)) = min(wid (k)), the hop count hdi (j) is set to 1. Therefore, the new calculation evaluates its hop count not from the relational distance of nodes i and j to the destination d, but rather from the immediate neighboring nodes, and its relation to shortest path. This leads to a perturbation based route discovery process that diverts depending on local conditions of neighboring node (e.g. node or link load).
Cloud computing requires fast discovery of services that are populated in large scale data centre networks. Once discovered, the paths for the services’ must also ensure that Quality-of-Service (QoS) and Quality-of-Experience (QoE) requirements are met. fs-PGBR provides individual routing paths that can adapt to each service’s requirements in a fully distributed manner, while minimizing the cost of discovery. In particular, fs-PGBR is ideal for cloud computing infrastructure that are built over large network topologies, and is able to support high service churn rates or varying user demand.
3.
5.
PRELIMINARY EVALUATION
We have evaluated and compared fs-PGBR to the original PGBR discovery mechanism, where simulations were conducted in our custom developed Java simulator. The simulation evaluated two performance metrics based on a realworld large scale topology (i.e. Sprintlink network). The weights that were used to calculate the shortest path were provided by the Rocketfuel project and the load of each link were uniformly selected between 0 and 100% (loads are not symmetric). The simulation scenario generated 100 sourcedestination pairs for each hop count distance, ranging from 1 to 7 hops. The simulations were reproduced 100 times with different loads, and source-destination pairs selection. For each simulation, we varied the value of γ from 0.1 to 1.0, which is the weighting for hdi (j) and reflects the importance of this weight with respect to the shortest path approach. In addition, α and β have similar weights, and set to make the sum of the three weights equal to 1. As shown in Figure 2, we evaluated the number of hops required to successfully reach the destination as well as the discovered path length ratio (Figure 2 (a) and (b)). We observed, in Figure 2 (a) and (b) that using the new hdi (j) calculation approach leads to minimal discovery time
ACKNOWLEDGEMENTS
This work has received support from Science Foundation Ireland via “A Biologically inspired framework supporting network management for the Future Internet” starting investigator award (grant no. 09/SIRG/I1643).
6.
REFERENCES
[1] S. Balasubramaniam, J. Mineraud, P. Mcdonagh, P. Perry, L. Murphy, W. Donnelly, and D. Botvich. An Evaluation of Parameterized Gradient Based Routing With QoE Monitoring for Multiple IPTV Providers. IEEE Transactions on Broadcasting, 57(2):183–194, 2011. [2] H. Ballani, P. Costa, T. Karagiannis, and A. Rowstron. Towards predictable datacenter networks. In Proceedings of the ACM SIGCOMM 2011 conference, SIGCOMM ’11, pages 242–253, New York, NY, USA, 2011. [3] H. T. Dinh, C. Lee, D. Niyato, and P. Wang. A survey of mobile cloud computing: architecture, applications, and approaches. Wireless Communications and Mobile Computing, accepted, 2011.
302
