# US 2023/0054873 A1 · Real time performance tuning of network devices

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2023/0054873 A1
- Document ID: US20230054873A1
- Status: Published application
- Kind: A1
- Application number: 17/981,516
- Filed: 2022-11-07
- Published: 2023-02-23
- Earliest claimed: 2020-12-22
- Assignees: Mellanox Technologies, Ltd.
- Applicants: Mellanox Technologies, Ltd.
- Inventors: Ido Gilboa; Shay Aisman; Sagi Arieli; Oren Vaserberger; Amit Mandelbaum; Doron Haritan Kazakov; Natali Shechtman; Iftah Levi; Amir Ancel
- Web page: https://patentgrep.com/patent/US20230054873A1

## CPC

- H04L 41/0836: {to enhance reliability, e.g. reduce downtime}
- H04L 41/0823: characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability (for optimising operational conditions of wireless networks H04W24/02)
- H04L 41/083: for increasing network speed
- H04L 41/085: Retrieval of network configuration; Tracking network configuration history
- H04L 41/0866: Checking the configuration
- H04L 41/0886: {Fully automatic configuration}
- H04L 43/08: Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
- H04L 47/283: in response to processing delays, e.g. caused by jitter or round trip time [RTT]
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)

## Abstract

A network device (ND) includes packet processing circuitry and performance optimization circuitry. The packet processing circuitry is connected to a network and is configured to process communication packets for communicating over the network. The packet processing circuitry includes a plurality of configuration registers for setting one or more operation parameters of the ND. The performance optimization circuitry is configured to improve a performance measure of the ND by iteratively calculating the performance measure and adjusting values of one or more of the configuration registers based on the performance measure.

## Claims

### Claim 1

A network device (ND), comprising:
packet processing circuitry, which is connected to a network and is configured to process communication packets for communicating over the network, the packet processing circuitry comprising a plurality of configuration registers for setting one or more operation parameters of the ND; and
performance optimization circuitry, configured to improve a performance measure of the ND by:
executing a genetic algorithm that continuously modifies candidate sets of the operation parameters;
calculating the performance measure based on the candidate sets of the operation parameters modified by the genetic algorithm; and
continuously adjusting values of one or more of the configuration registers based on the performance measure.

### Claim 2 (depends on 1)

The ND according to claim 1, wherein the optimization circuitry is configured to adjust the values of the one or more of the configuration registers responsively to a parameter ruleset, which sets one or more constraints on the one or more of the operation parameters.

### Claim 3 (depends on 1)

The ND according to claim 1, wherein the optimization circuitry is configured to adjust the values of the one or more of the configuration registers responsively to a cross-parameters ruleset, which sets one or more constraints on inter-dependency between at least two of the operation parameters.

### Claim 4 (depends on 1)

The ND according to claim 1, wherein the optimization circuitry is configured to improve the performance measure of the ND by:
selecting an initial group of candidate vectors, each vector specifying the values for a respective subset of the configuration registers; and
modifying the candidate vectors in an iterative process, by adjusting the values of the one or more configuration registers responsively to the performance measure corresponding to the candidate vectors.

### Claim 5 (depends on 4)

The ND according to claim 4, wherein the optimization circuitry is configured to:
store a value of the performance measure at a first stage of the iterative process;
at a second, later stage of the iterative process, assess an extent of change of the performance metric relative to the stored value; and
refresh one or more of the candidate vectors responsively to the value exceeding a threshold.

### Claim 6 (depends on 1)

The ND according to claim 1, wherein the optimization circuitry is configured to improve the performance measure of the ND by, using the genetic algorithm, iteratively modifying a group of candidate vectors, each vector specifying a respective subset of the configuration registers, wherein the genetic algorithm comprises: i) a parents selection step, in which pairs of candidate vectors are selected, ii) a cross-over step, in which child vectors are generated by mixing values of pairs of candidate vectors, iii) a mutation step, in which mutated child vectors are generated by modifying the child vectors and, iv) a fittest-selection step, in which candidate vectors for an iteration are selected responsively to the performance measure of the mutated child vectors of a previous iteration.

### Claim 7 (depends on 1)

The ND according to claim 1, wherein the optimization circuitry is configured to improve the performance measure of the ND by, using the genetic algorithm, iteratively modifying a group of candidate vectors, each vector specifying a respective subset of the configuration registers, wherein the genetic algorithm comprises: i) a mutation step, in which candidate vectors are modified, and ii) a cross-over step, in which child vectors are generated by mixing values of pairs of candidate vectors.

### Claim 8 (depends on 1)

The ND according to claim 1, wherein the performance measure is indicative of a receive bandwidth.

### Claim 9 (depends on 1)

The ND according to claim 1, wherein the performance measure is indicative of a transmit bandwidth.

### Claim 10 (depends on 1)

The ND according to claim 1, wherein the performance measure is indicative of a packet drop rate.

### Claim 11 (depends on 1)

The ND according to claim 1, wherein the performance measure is indicative of a packet pause rate.

### Claim 12 (depends on 1)

The ND according to claim 1, wherein the performance measure is indicative of a combination of latency and memory utilization.

### Claim 13 (depends on 1)

The ND according to claim 1, wherein the performance measure is indicative of a bit-error rate (BER).

### Claim 14

A method in a network device (ND) connected to a network, the method comprising:
using packet processing circuitry in the ND, processing communication packets for communicating over the network, wherein the packet processing circuitry comprises a plurality of configuration registers for setting one or more operation parameters of the ND; and
using performance optimization circuitry in the ND, improving a performance measure of the ND by:
executing a genetic algorithm that continuously modifies candidate sets of the operation parameters;
calculating the performance measure based on the candidate sets of the operation parameters modified by the genetic algorithm; and
continuously adjusting values of one or more of the configuration registers based on the performance measure.

### Claim 15 (depends on 14)

The method according to claim 14, wherein adjusting the values of the one or more of the configuration registers is performed responsively to a parameter ruleset, which sets one or more constraints on the one or more of the operation parameters.

### Claim 16 (depends on 14)

The method according to claim 14, wherein adjusting the values of the one or more of the configuration registers is performed responsively to a cross-parameters ruleset, which sets one or more constraints on inter-dependency between at least two of the operation parameters.

### Claim 17 (depends on 14)

The method according to claim 14, wherein improving the performance measure of the ND comprises:
selecting an initial group of candidate vectors, each vector specifying the values for a respective subset of the configuration registers; and
modifying the candidate vectors in an iterative process, by adjusting the values of the one or more configuration registers responsively to the performance measure corresponding to the candidate vectors.

### Claim 18 (depends on 17)

The method according to claim 17, wherein improving the performance measure of the ND comprises:
storing a value of the performance measure at a first stage of the iterative process;
at a second, later stage of the iterative process, assessing an extent of change of the performance metric relative to the stored value; and
refreshing one or more of the candidate vectors responsively to the value exceeding a threshold.

### Claim 19 (depends on 14)

The method according to claim 14, wherein improving the performance measure of the ND comprises, using the genetic algorithm, iteratively modifying a group of candidate vectors, each vector specifying a respective subset of the configuration registers, wherein the genetic algorithm comprises: i) a parents selection step, in which pairs of candidate vectors are selected, ii) a cross-over step, in which child vectors are generated by mixing values of pairs of candidate vectors, iii) a mutation step, in which mutated child vectors are generated by modifying the child vectors and, iv) a fittest-selection step, in which candidate vectors for an iteration are selected responsively to the performance measure of the mutated child vectors of a previous iteration.

### Claim 20 (depends on 14)

The method according to claim 14, wherein improving the performance measure of the ND comprises, using the genetic algorithm, iteratively modifying a group of candidate vectors, each vector specifying a respective subset of the configuration registers, wherein the genetic algorithm comprises: i) a mutation step, in which candidate vectors are modified, and ii) a cross-over step, in which child vectors are generated by mixing values of pairs of candidate vectors.
