公开(公告)号：US11522780B1

公开(公告)日：2022-12-06

申请号：US17118226

申请日：2020-12-10

Applicant: Amazon Technologies, Inc.

Inventor： Ronan James Wallace ,  Edin Kozo Becirevic

IPC: H04L43/12 ,  H04L43/04 ,  H04L43/062 ,  H04L43/0852 ,  H04L43/0823 ,  H04L43/0876 ,  H04L43/50 ,  H04L43/0817

Abstract: Usually agents probe other agents across the network to determine health of a network. However, it is also useful for agents to probe themselves. Such a test probe is off-network in that it does not leave a server rack. The off-network probes can be used to detect noisy agents, which can be excluded from further consideration in evaluating overall network quality. The off-network probing of the agents can be done in a loopback interface of the host device or via a top-of-rack switch to another distinct agent in the rack.

公开(公告)号：US12010007B1

公开(公告)日：2024-06-11

申请号：US17203376

申请日：2021-03-16

Applicant: Amazon Technologies, Inc.

Inventor： Ronan James Wallace

IPC: G06F15/16 ,  H04L43/0852 ,  H04L43/12 ,  H04L43/16

CPC classification number: H04L43/16 ,  H04L43/0852 ,  H04L43/12

Abstract: Usually agents probe other agents across the network to determine the network health. If network data from noisy agents (i.e., agents themselves causing the high latency) is used, it degrades the overall probe statistics. Thus, agents can be identified as noisy and then data from such noisy agents can be excluded from further consideration, thereby improving the quality of the active monitoring dataset. An algorithm executing on a centralized server computer uses a sample of active monitoring probe data. A first part of the algorithm involves determining the expected latency for a probe. The second part of the algorithm uses the information about the expected latency for a probe to compute a fraction of an agent's probes that have anomalous latencies using the original (full) data set. Agents with large fractions of anomalous probes receive high noise scores in the algorithm and are filtered.