Memory mode categorizations
AssigneeHewlett-Packard Development Company, L.P.
InventorsRaphael GAY, Kirsten OLSEN
Life of this patent
Estimating the term…
Abstract
Example implementations relate to memory mode categorization. An example memory mode categorization can include determining local and remote data bandwidths received at each of a first processor and a second processor for a data sample, comparing the local and the remote data bandwidths to a first threshold bandwidth and a second threshold bandwidth, respectively, creating a traffic pattern for the data sample based on the comparison, and categorizing the data sample as being a candidate for a particular memory mode based on the created traffic pattern.
Claims15
A machine-readable medium storing instructions executable by a processing resource to cause a computing system to:
- determine local and remote data bandwidths received at each of a first processor and a second processor for a data sample;
- compare the local and the remote data bandwidths to a first threshold bandwidth and a second threshold bandwidth, respectively;
- create a traffic pattern for the data sample based on the comparison;
- and
- categorize the data sample as being a candidate for a particular memory mode based on the created traffic pattern.
A system, comprising:
- a sample engine to collect for a plurality of data samples associated with a running workload, local and remote data bandwidth values received at each of a first processor and a second processor;
- a pattern engine to construct traffic patterns for each of the plurality of data samples based on the received local and remote data bandwidth values;
- a categorization engine to categorize each of the constructed traffic patterns as memory node interleaved (MNI) memory mode-favorable, non-uniform memory access (NUMA) memory mode-favorable, or indifferent based on the constructed pattern;
- and
- a recommendation engine to determine a favorable memory mode for the running workload based on the categorized constructed traffic patterns.
A method, comprising,
- calculating local and remote data bandwidths across a plurality of processors for a plurality of data samples;
- comparing each of the local data bandwidths to a local data bandwidth threshold;
- comparing each of the remote data bandwidths to a remote data bandwidth threshold;
- constructing a traffic pattern for each of the plurality of data samples based on the comparisons;
- detecting memory node interleaved (MNI) memory mode-favorable data samples from within the plurality of data samples based on the constructed traffic patterns;
- labeling remaining data samples as non-uniform memory access (NUMA) memory mode-favorable or indifferent data samples based on the detection;
- calculating an index value by assigning a weight to each of the remote data bandwidths and calculating a difference with a respective local memory bandwidth;
- and
- categorizing the plurality of data samples based on the detected MNI-favorable data samples, the NUMA or indifferent labels, and the calculated index values.
Dependencies are inferred from the claim text.