BID-Scheme
It is an entry after a very long time. After my internship at INRIA, France, my research direction suddenly changed into bandwidth based resource discovery. As the first step towards it, I have been working on a landmarking scheme for distributed bandwidth prediction.
The bandwidth landmarking scheme I developed, which is named as BID-scheme, in inspired from landmarking schemes for network latency prediction and another bandwidth prediction scheme called BRoute. I will make another detailed entry on BRoute. I build the node states for bandwidth prediction similar to the latency-based landmarking scheme, but uses a deduction scheme similar to the one used in BRoute.
The landmarks in my BID-scheme are not some chosen nodes as in the latency-based landmarking schemes. BID-scheme uses some selected autonomous system (AS) domains as landmark domains (not landmark nodes). Another difference between the BID-scheme and the latency-based landmarking schemes are that, in BID-scheme, nodes are not placed in a Cartesian space -- it is impossible because bandwidth is not an additive measure like latency (and neither abides by the triangular inequality). Instead, the nodes in the BID-scheme probes the landmark domains to construct BID tables. Each row in a BID table belongs to a landmark AS and records information about it -- distance to it in terms of AS hop counts, average bottleneck bandwidth observed in reaching the AS and a bloom filter that stores information about the landmark AS's probability of being in the route towards other ASes.
Deduction of bandwidth between two end-points works in two phases: first to guess the most probable AS domain (from the landmark set) to be in between the two end-points; and then to deduce the end-to-end bandwidth using the two links -- source point to middle AS and middle AS to sink point. The bloom filters from the BID-table from both end-nodes are used to determine the best probable middle AS and the respective bandwidths are used to determine the end-to-end bandwidth.
One definite advantage of the BID-scheme over the BRoute scheme is that the BID-tables of the nodes can be used to create bandwidth identifiers (BIDs) of the nodes which can be used to create BID-based routing, in other words, a bandwidth-based discovery scheme. This BID-based routing/discovery scheme is a direct derivative of the DHT-based discovery schemes like Pastry.
I performed a simulation study on the scheme which showed promising results. We have submitted a paper to GlobeCom 2007 on this research. A journal version is currently under process.
The bandwidth landmarking scheme I developed, which is named as BID-scheme, in inspired from landmarking schemes for network latency prediction and another bandwidth prediction scheme called BRoute. I will make another detailed entry on BRoute. I build the node states for bandwidth prediction similar to the latency-based landmarking scheme, but uses a deduction scheme similar to the one used in BRoute.
The landmarks in my BID-scheme are not some chosen nodes as in the latency-based landmarking schemes. BID-scheme uses some selected autonomous system (AS) domains as landmark domains (not landmark nodes). Another difference between the BID-scheme and the latency-based landmarking schemes are that, in BID-scheme, nodes are not placed in a Cartesian space -- it is impossible because bandwidth is not an additive measure like latency (and neither abides by the triangular inequality). Instead, the nodes in the BID-scheme probes the landmark domains to construct BID tables. Each row in a BID table belongs to a landmark AS and records information about it -- distance to it in terms of AS hop counts, average bottleneck bandwidth observed in reaching the AS and a bloom filter that stores information about the landmark AS's probability of being in the route towards other ASes.
Deduction of bandwidth between two end-points works in two phases: first to guess the most probable AS domain (from the landmark set) to be in between the two end-points; and then to deduce the end-to-end bandwidth using the two links -- source point to middle AS and middle AS to sink point. The bloom filters from the BID-table from both end-nodes are used to determine the best probable middle AS and the respective bandwidths are used to determine the end-to-end bandwidth.
One definite advantage of the BID-scheme over the BRoute scheme is that the BID-tables of the nodes can be used to create bandwidth identifiers (BIDs) of the nodes which can be used to create BID-based routing, in other words, a bandwidth-based discovery scheme. This BID-based routing/discovery scheme is a direct derivative of the DHT-based discovery schemes like Pastry.
I performed a simulation study on the scheme which showed promising results. We have submitted a paper to GlobeCom 2007 on this research. A journal version is currently under process.
posted by Unknown at
11:31 PM
0 Comments:
Post a Comment
<< Home