Client-space network monitoring
A method is provided for client-side monitoring in a cloud provider environment including a network having a plurality of nodes. The method includes estimating a logical topology of the network by clustering source-destination node pairs from among the plurality of nodes based on end-to-end performance measurements between various ones of the plurality of nodes. The method further includes estimating a performance state of the network for a client application based on end-to-end performance measurements. The method also includes at least one of scheduling and customizing a workload within the network based on the estimated logical topology and the estimated performance state of the network.
1. A computer-implemented method for client-side monitoring in a cloud provider environment comprising a network having a plurality of nodes, the method comprising:
estimating a logical topology of the network by clustering source-destination node pairs while excluding intermediate nodes from among the plurality of nodes based on end-to-end performance measurements between various ones of the plurality of nodes forming the source-destination node pairs;
estimating a performance state of the network for a client application based on end-to-end performance measurements; and
at least one of scheduling and customizing a workload within the network, and executing the workload within the network, based on the estimated logical topology and the estimated performance state of the network.
2. The computer-implemented method of claim 1 , wherein estimating the logical topology of the network comprises:
obtaining a targeted network architecture for the network;
clustering the source-destination pairs formed by the plurality of nodes into at least two clusters based on end-to-end performance measurements between the source nodes and the destination nodes;
estimating a minimum logical network topology and assigning the source nodes and the destination nodes from the pairs to nodes in the minimum logical topology based on clustering results from the end-to-end performance measurements; and
augmenting the minimum logical topology to satisfy the targeted network architecture.
3. The computer-implemented method of claim 2 , wherein said obtaining step comprises obtaining multiple targeted architectures, each with different numbers of levels, to obtain a multi-resolution topology inference.
4. The computer-implemented method of claim 1 , wherein estimating the performance state of the network comprises:
identifying a performance-measurement-to-link relationship between end-to-end performance measurements and link performances, the end-to-end performance measurements being for various pairings of the plurality of nodes;
computing minimum link groups whose aggregate performances can be uniquely identified from the end-to-end performance measurements based on the relationship;
computing weights for each of the minimum link groups based on application-specified per-link weights;
collecting the end-to-end performance measurements between the various pairings of the plurality of nodes; and
inferring an aggregate performance state of each of the minimum link groups.
5. A non-transitory article of manufacture tangibly embodying a computer readable program which when executed causes a computer to perform the steps of claim 1 .
6. The computer-implemented method of claim 1 , wherein the logical topology is estimated to have logical distances in the logical topology that maximally correlate with the end-to-end performance measurements such that a given one of the source-destination node pairs having a better connection than another one of the source-destination node pairs has a given constituent source node closer to a given constituent destination node in the logical topology than the other one of the source-destination node pairs.