TY - GEN
T1 - Multiple graph abstractions for parallel routing over virtual topologies
AU - Soran, Ahmet
AU - Yuksel, Murat
AU - Gunes, Mehmet Hadi
N1 - Publisher Copyright:
© 2017 IEEE.
PY - 2017/11/20
Y1 - 2017/11/20
N2 - High throughput data transfers across the Internet has become a challenge with deployment of data centers and cloud platforms. In this paper, we propose to utilize the cores of a router to build multiple abstractions of the underlying topology to parallelize end-to-end (e2e) streams for bulk data transfers. By abstracting a different graph for each core, we steer each core to calculate a different e2e path in parallel. The e2e transfers can use the shortest paths obtained from each subgraph to increase the total throughput over the underlying network. Even though calculating shortest paths is well optimized in legacy routing protocols (e.g., OSPF), finding optimal set of subgraphs to generate non-overlapping and effective multiple paths is a challenging problem. To this end, we analyze centrality metrics to eliminate potentially highest loaded routers or edges in the topology without coordination and eliminate them from the subgraphs. We evaluate the heuristics in terms of aggregate throughput and robustness against failures.
AB - High throughput data transfers across the Internet has become a challenge with deployment of data centers and cloud platforms. In this paper, we propose to utilize the cores of a router to build multiple abstractions of the underlying topology to parallelize end-to-end (e2e) streams for bulk data transfers. By abstracting a different graph for each core, we steer each core to calculate a different e2e path in parallel. The e2e transfers can use the shortest paths obtained from each subgraph to increase the total throughput over the underlying network. Even though calculating shortest paths is well optimized in legacy routing protocols (e.g., OSPF), finding optimal set of subgraphs to generate non-overlapping and effective multiple paths is a challenging problem. To this end, we analyze centrality metrics to eliminate potentially highest loaded routers or edges in the topology without coordination and eliminate them from the subgraphs. We evaluate the heuristics in terms of aggregate throughput and robustness against failures.
UR - http://www.scopus.com/inward/record.url?scp=85041289298&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85041289298&partnerID=8YFLogxK
U2 - 10.1109/INFCOMW.2017.8116496
DO - 10.1109/INFCOMW.2017.8116496
M3 - Conference contribution
AN - SCOPUS:85041289298
T3 - 2017 IEEE Conference on Computer Communications Workshops, INFOCOM WKSHPS 2017
SP - 904
EP - 909
BT - 2017 IEEE Conference on Computer Communications Workshops, INFOCOM WKSHPS 2017
T2 - 2017 IEEE Conference on Computer Communications Workshops, INFOCOM WKSHPS 2017
Y2 - 1 May 2017 through 4 May 2017
ER -