Back to results
US 12,255,670 B2Granted
Compression strategy selection powered by machine learning
AssigneeMellanox Technologies, Ltd.
InventorsVladimir Shalikashvili, Ran Sandhaus
Life of this patent
Estimating the term…
Abstract
A data compression system comprising computer memory to store plural compression algorithms and a hardware processor to apply compression algorithm/s to incoming data items, wherein the compression algorithm to be applied to individual data item/s from among the incoming data items is selected, from among the plural compression algorithms, by the hardware processor, depending at least on the individual data item.
Claims28
A data compression system comprising:
- a computer memory to store plural compression algorithms;
- and
- a hardware processor to apply at least one picked compression algorithm to incoming data items, wherein the picked compression algorithm to be applied to at least one individual data item from among the incoming data items is selected, from among the plural compression algorithms, by the hardware processor, depending at least on said individual data item,
- wherein at least the processor is deployed in a communication system including at least one transmitter and at least one receiver, thereby to define a transmitter side of the system and a receiver side of the system, and
- wherein the transmitter side of the system is implemented by a dedicated hardware accelerator,
- wherein the hardware processor selects the picked compression algorithm using a machine learning model and employs a best-ness criterion which is a function of plural characteristics of each compression algorithm and wherein the plural characteristics are all normalized to a single range and then combined using a combination function.
3 more elements
A data compression system comprising:
- a computer memory to store plural compression algorithms;
- and
- a hardware processor to apply at least one picked compression algorithm to incoming data items, wherein the picked compression algorithm to be applied to at least one individual data item from among the incoming data items is selected, from among the plural compression algorithms, by the hardware processor, depending at least on said individual data item,
- wherein the computer memory is to store the plural compression algorithms in association with respective plural data-item classes, and wherein a machine learning model is used to select the picked compression algorithm from among the plural compression algorithms by using the hardware processor, for at least one individual data item from among the incoming data items, to classify said individual data item as belonging to a class C from among the plural data-item classes and to select a compression algorithm associated with class C in said computer memory, as the picked compression algorithm for said individual data item.
1 more element
A data compression system comprising:
- a computer memory to store plural compression algorithms;
- and
- a hardware processor to apply at least one picked compression algorithm to incoming data items, wherein the picked compression algorithm to be applied to at least one individual data item from among the incoming data items is selected, from among the plural compression algorithms, by the hardware processor, depending at least on said individual data item,
- wherein the processor is configured in software to select and apply the picked compression algorithm to incoming data items and wherein the plural compression algorithms are offloaded to hardware, and
- wherein the plural compression algorithms are offloaded to a GPU, and
- wherein the processor is deployed in a data center networking management platform with network telemetry functionality and wherein said incoming data items store network telemetry data.
3 more elements
A data compression method comprising using a hardware processor for:
- predicting a best compression algorithm for each of plural incoming data items, thereby to define plural best compression algorithms for said plural items respectively;
- performing at least one of the following operations, A and B:
- Operation A:
- for each data item A from among the plural items, which is to be compressed, retrieving compression code corresponding to a best algorithm A defined for item A by said predicting, and compressing said item A by running said compression code on said item A;
- and/or
- Operation B:
- for each data item B from among the plural items, which is compressed, retrieving decompression code corresponding to the best algorithm B defined for item B by said predicting, and decompressing said item B by running said decompression code on said item B;
- and
- using a hardware processor for performing, offline, the following:
- creating a dataset with plural data entries, each including a data item paired with an indication of a best compression algorithm for that data item;
- training a model to predict a best compression algorithm for each incoming data item, using said dataset as training data, thereby to provide a trained model;
- and
- deploying said trained model.
10 more elements
Dependencies are inferred from the claim text.