Machine learning approach for estimating a network path property

1. A method of estimating a network path property between two nodes in a network, the method comprising:

receiving network path property measurements for paths between each of the two nodes and a subset of nodes in the network;

determining, by a computer system, node profiles for the two nodes using the received network path property measurements, wherein the node profiles are used to estimate the network path property between the two nodes, and wherein determining node profiles comprises determining network path property vectors including the received network path property measurements for the two nodes, and truncating the vectors by a predetermined number of values; and

estimating the network path property between the two nodes based on the node profiles,

wherein the subset of nodes comprises milestone nodes, and wherein determining node profiles for the two nodes using the received network path property measurements comprises:

determining common nodes that are common for upstream and downstream routes, wherein the upstream routes are between one of the two nodes and the milestone nodes and the downstream routes are between the milestone nodes and the other one of the two nodes;

for every common node,

determining a first network path property vector including measurements for one of the upstream routes between one of the two nodes and the common node; and

determining a second network path property vector including measurements for one of the downstream routes between the other one of the two nodes and the common node.

2. The method of claim 1 , wherein estimating comprises:

using machine learning to determine probabilities for possible estimations of the network path property; and

selecting one of the possible estimations as the estimated network path property.

3. The method of claim 1 , further comprising:

determining a count of a number of the measurements in the vectors for the nodes that are the same; and

including the count in the node profiles.

4. The method of claim 1 , further comprising:

determining a histogram for each of the vectors, wherein the histograms are used in the node profiles.

5. The method of claim 1 , wherein the subset of nodes comprises milestone nodes, and receiving network path property measurements for paths between each of the two nodes and a subset of nodes in the network comprises:

receiving network path property measurements for paths between each of the two nodes and landmark nodes of the milestone nodes, and paths between each of the two nodes and intermediate routers of the milestone nodes.

6. The method of claim 1 , wherein using machine learning to estimate the network path property between the two nodes based on the node profiles further comprises:

using a Bayesian trained classifier to estimate the network path property.

7. The method of claim 1 , wherein the network path property comprises one of hop count, latency, bandwidth and loss.

8. A method of estimating a network path property between two nodes in a network, the method comprising:

receiving network path property measurements for paths between each of the two nodes and a subset of nodes in the network, wherein the subset of nodes comprises milestone nodes; and

determining, by a computer system, node profiles for the two nodes using the received network path property measurements by

determining common nodes that are common for upstream and downstream routes, wherein the upstream routes are between one of the two nodes and the milestone nodes and the downstream routes are between the milestone nodes and the other one of the two nodes;

for every common node,

determining a first network path property vector including measurements for one of the upstream routes between one of the two nodes and the common node;

determining a second network path property vector including measurements for one of the downstream routes between the other one of the two nodes and the common node; and

determining a histogram for the first and second vectors; and

using machine learning to estimate the network path property between the two nodes based on the node profiles.

9. A machine learning prediction system having a processor operable to estimate a network path property between two nodes in a network, the system comprising:

a profiler operable to receive network path property measurements for paths between each of the two nodes and milestone nodes in the network, and determine profiles for each of the two nodes based on the received network path property measurements, wherein the profiles are used to estimate the network path property between the two nodes and wherein the profiles are determined by determining network path property vectors including the measurements for the two nodes, and truncating the vectors by a predetermined number of values; and

a prediction engine using machine learning technology to estimate the network path property between the two nodes based on the determined profiles, wherein the machine learning technology includes using machine learning to determine probabilities for possible estimations of the network path property; and selecting one of the possible estimations as the estimated network path property,

wherein the profiler is operable to determine node profiles for the two nodes using the received network path property measurements by:

determining common nodes that are common for upstream and downstream routes, wherein the upstream routes are between one of the two nodes and the milestone nodes and the downstream routes are between the milestone nodes and the other one of the two nodes;

for every common node,

determining a first network path property vector including measurements for one of the upstream routes between one of the two nodes and the common node;

determining a second network path property vector including measurements for one of the downstream routes between the other one of the two nodes and the common node; and

determining a histogram for the first and second vectors.

10. The system of claim 9 , wherein the network path property comprises one of hop count, latency, loss and bandwidth.

11. The system of claim 9 , wherein the milestone nodes comprise landmark nodes in the network and intermediate routers in the network, wherein an intermediate router is a node encountered in an upstream route from a first node of the two nodes to a landmark node and encountered in a downstream route from a landmark node to a second node of the two nodes; and

the measurements comprise measurements to the milestone nodes.

12. The system of claim 9 , wherein the machine learning technology comprises a Bayesian trained classifier.

13. The system of claim 9 , wherein the profiler is operable to determine a count of a number of the measurements in the vectors that are the same, and include the count in the node profiles.

14. The system of claim 11 , wherein the profiler is operable to determine milestone node histograms from the vectors, wherein the milestone node histograms are used to determine the node profiles and are determined for each common milestone node in upstream routes and downstream routes for the nodes.

15. The system of claim 9 , wherein the profiler is operable to determine node histograms from the vectors, wherein the node histograms are used to determine the profiles.

16. Computer software embedded on a non-transitory computer readable medium, the computer software including instructions when executed performing a method of estimating a network path property , the method comprising:

receiving network path property measurements for paths between each node of a first set of nodes and milestone nodes in a network;

determining a node profile for each node of the first set of nodes by determining network path property vectors and truncating the vectors by a predetermined number of values, wherein each node profile includes the received network path property measurements between the node and the milestone nodes; and

using machine learning to estimate a network path property between two nodes in the first set, wherein the machine learning estimation of the network path property uses the node profiles for the two nodes,

wherein determining a node profile for each node of the first set of nodes comprises:

determining milestone nodes that are common for upstream and downstream routes, wherein the upstream routes are between one of the two nodes and the milestone nodes and the downstream routes are between the milestone nodes and the other one of the two nodes;

for every common milestone node,

determining a first network path property vector including measurements for one of the upstream routes between one of the two nodes and the common milestone node;

determining a second network path property vector including measurements for one of the downstream routes between the other one of the two nodes and the common milestone node; and

determining a histogram for the first and second vectors.

17. The computer software embedded on a non-transitory computer readable medium of claim 16 , wherein the method further comprises:

using the machine learning to determine probabilities for possible estimations of the network path property between the two nodes; and

selecting one of the possible estimations as the estimated network path property.