Workload summarization for congestion avoidance in computer servers
AssigneeInternational Business Machines Corporation
InventorsAlexei Karve, Maroun Touma, Kugamoorthy Gajananan, Sekou Lionel Remy
Life of this patent
Estimating the term…
Abstract
Methods, systems and computer program products for workload summarization and congestion avoidance are provided. Aspects include generating a workload summary associated with the workload request responsive to receiving a workload request. Aspects also include detecting potential congestion associated with the workload request by predicting resource usage based on the workload summary and a knowledge base model. Aspects further include performing an action to avoid the potential congestion, the action comprising holding off the workload request for at least a predetermined duration.
Claims14
A computer-implemented method for workload summarization and congestion avoidance, the computer-implemented method comprising:
- responsive to receiving a workload request, obtaining a packet capture (Pcap) file associated with the workload request and generating a workload summary associated with the workload request by automatically parsing the Pcap file to extract, for each protocol present, a count of frames and a count of bytes;
- detecting potential congestion associated with the workload request by inputting the extracted per-protocol frame and byte counts as numeric features into a trained machine-learning model resident in memory to obtain a quantitative prediction of resource usage, including at least one of CPU, memory, or bandwidth, and predicting resource usage based on the workload summary and a knowledge base model;
- and
- performing an action to avoid the potential congestion, the action comprising writing the workload request into a delayed-execution queue to hold off the workload request for at least a predetermined duration, wherein holding off prevents pod eviction in a cluster environment.
A system comprising:
- a memory;
- and
- one or more processing units coupled with the memory, the one or more processing units configured to perform a method for workload summarization and congestion avoidance, the method comprising:
- responsive to receiving a workload request, obtaining a packet capture (Pcap) file associated with the workload request and generating a workload summary associated with the workload request by automatically parsing the Pcap file to extract, for each protocol present, a count of frames and a count of bytes;
- detecting potential congestion associated with the workload request by inputting the extracted per-protocol frame and byte counts as numeric features into a trained machine-learning model resident in memory to obtain a quantitative prediction of resource usage, including at least one of CPU, memory, or bandwidth, and predicting resource usage based on the workload summary and a knowledge base model;
- and
- performing an action to avoid the potential congestion, the action comprising writing the workload request into a delayed-execution queue to hold off the workload request for at least a predetermined duration, wherein holding off prevents pod eviction in a cluster environment.
A computer program product comprising a memory with one or more computer executable instructions stored thereon, which when executed by one or more processing units cause the one or more processing units to perform a method for workload summarization and congestion avoidance, the method comprising:
- responsive to receiving a workload request, obtaining a packet capture (Pcap) file associated with the workload request and generating a workload summary associated with the workload request by automatically parsing the Pcap file to extract, for each protocol present, a count of frames and a count of bytes;
- detecting potential congestion associated with the workload request by inputting the extracted per-protocol frame and byte counts as numeric features into a trained machine-learning model resident in memory to obtain a quantitative prediction of resource usage, including at least one of CPU, memory, or bandwidth, and predicting resource usage based on the workload summary and a knowledge base model;
- and
- performing an action to avoid the potential congestion, the action writing the workload request into a delayed-execution queue to hold off the workload request for at least a predetermined duration, wherein holding off prevents pod eviction in a cluster environment.
Dependencies are inferred from the claim text.