# US 2022/0011966 A1 · Reduced network load with combined put or get and receiver-managed offset

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

- Number: US 2022/0011966 A1
- Document ID: US20220011966A1
- Status: Published application
- Kind: A1
- Application number: 17/485,114
- Filed: 2021-09-24
- Published: 2022-01-13
- Earliest claimed: 2021-09-24
- Assignees: Intel Corporation
- Applicants: Intel Corporation
- Inventors: David KEPPEL; David M. OZOG
- Web page: https://patentgrep.com/patent/US20220011966A1

## CPC

- G06F 3/061: {Improving I/O performance}
- G06F 3/0655: {Vertical data movement, i.e. input-output transfer; data movement between one or more hosts and one or more storage devices}
- G06F 3/0604: {Improving or facilitating administration, e.g. storage management}
- G06F 3/0659: {Command handling arrangements, e.g. command buffers, queues, command scheduling}
- G06F 3/067: {Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Methods and apparatus for reduced network load with receiver-managed offset (RMO) PUT or GET messages. An RMO PUT message including an RMO key, data, and a length is sent from an initiator to a target, where the RMO key is extracted by a Network Interface controller (NIC), SmartNIC, or Infrastructure Processing Unit and used to identify an address or address offset of a memory buffer in a target memory at which to write the data. An RMO GET message is sent from an initiator to a target and includes an RMO key, a source buffer on the target, and a length. The target processes the RMO GET, reads the length of data from its source buffer, and returns a message to the initiator including the RMO key, the read data, and the length. The RMO key is extracted and used to identify an address or address offset of a memory buffer in a memory on the initiator in which to write the read data.

## Claims

### Claim 1

A method comprising:
sending a first message comprising a receiver-managed offset (RMO) PUT or UPDATE from an initiator to a target having a plurality of processing elements (PEs) and one or more memories operatively coupled to a Network Interface controller (NIC), the message including an RMO key, data, and a length;
upon receipt of the first message at the NIC on the target, extracting the RMO key and employing the RMO key to identify an address or address offset of a memory buffer in a memory; and
one of,
writing, via the NIC, the data at the address or address offset in the memory buffer when the first message is an RMO PUT; otherwise
applying an update operation, via the NIC, to existing data at the address or address offset in the memory buffer using the data in an RMO UPDATE.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
implementing a list or table on the target comprising a plurality of indicators indexed by an index value;
one of,
i. computing an index value using the RMO key as an input to the key function; or
ii. employing the RMO key as an index value;
identifying an indicator in the list or table using the index value; and
employing the indicator to determine the address or address offset in the memory buffer.

### Claim 3 (depends on 2)

The method of claim 2, wherein the RMO key comprises a PE number comprising a first integer and an RMO number comprising a second integer, and where the key function comprises a bitwise concatenation of the PE number and the RMO number or a bitwise concatenation of the RMO number and PE number.

### Claim 4 (depends on 2)

The method of claim 2, wherein the indicator comprises one of a counter or a pointer to a counter, further comprising:
using a current counter value to determine an address offset in the memory buffer at which to write the data; and
one of,
i. updating the current counter value as a function of the length prior to writing the data; or
ii. updating the current counter value as a function of the length after writing the data.

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
implementing an associative key-counter table on the target;
employing the RMO key as an input to identify a counter; and
employing the counter to determine an offset address at which to write the data.

### Claim 6 (depends on 1)

The method of claim 1, wherein the first message further includes a PE tag comprising one of a node identifier and a PE identifier or a PE identifier.

### Claim 7 (depends on 1)

The method of claim 1, further comprising:
sending a second message comprising an RMO GET message from the initiator to the target, the RMO GET message including an RMO key, a destination buffer, and a length;
processing, via the NIC on the target, the RMO GET message to extract the RMO key and employ the RMO key to identify an address or address offset of a memory buffer in a memory;
read, via the NIC, data having the length beginning at the address or address offset in the memory buffer;
returning a third message from the target to the initiator including a data payload comprising the data that are read and the destination buffer; and
writing, at the initiator, the data payload into the destination buffer.

### Claim 8 (depends on 1)

The method of claim 1, wherein the initiator has a plurality of processing elements (PEs) and one or more memories operatively coupled to a NIC, further comprising:
sending a second message comprising an RMO GET message from the initiator to the target, the RMO GET message including an RMO key, a source buffer, and a length;
processing, via the NIC on the target, the RMO GET message to read data having the length from the source buffer;
returning a third message from the target to the initiator including the RMO key, a data payload comprising the data that are read, and the length;
upon receipt of the third message at the NIC on the initiator, extracting the RMO key and employing the RMO key to identify an address or address offset of a memory buffer in a memory; and
writing, via the NIC on the initiator, the data payload at the address or address offset in the memory buffer.

### Claim 9

An apparatus, configured to be implemented in a compute node including a plurality of processing elements (PEs) and one or more memories operatively coupled to the apparatus when installed in the compute node, comprising:
one or more network ports; and
circuitry and logic to:
receive, from an initiator at a network port, a first message comprising a receiver-managed offset (RMO) PUT or UPDATE and including an RMO key, data, and a length;
extract the RMO key from the first message and employ the RMO key to identify an address or address offset of a memory buffer in a memory; and
one of,
write the data to the memory buffer beginning at the address or address offset in the memory buffer when the first message is an RMO PUT; otherwise
apply an update operation, via the NIC, to existing data at the address or address offset in the memory buffer using the data in an RMO UPDATE.

### Claim 10 (depends on 9)

The apparatus of claim 9, further comprising circuitry and logic to:
implement a list or table comprising a plurality of indicators indexed by an index value;
one of,
i. compute an index value using the RMO key as an input to a key function; or
ii. employ the RMO key as an index value;
identify an indicator in the list or table using the index value; and
employ the indicator to determine the address or address offset in the memory buffer.

### Claim 11 (depends on 10)

The apparatus of claim 10, wherein the RMO key comprises a PE number comprising a first integer and an RMO number comprising a second integer, and where the key function comprises a bitwise concatenation of the PE number and the RMO number or a bitwise concatenation of the RMO number and PE number.

### Claim 12 (depends on 9)

The apparatus of claim 9, wherein the indicator comprises one of a counter or a pointer to a counter, further comprising circuitry and logic to:
use a current counter value to determine an address offset in the memory buffer at which to write the data; and
one of,
i. update the current counter value as a function of the length prior to writing the data; or
ii. update the current counter value as a function of the length after writing the data.

### Claim 13 (depends on 9)

The apparatus of claim 9, further comprising circuitry and logic to:
implement an associative key-counter table on the target;
employ the RMO key as an input to the key-counter table to identify a counter; and
employ the counter to determine an offset address at which to write the data.

### Claim 14 (depends on 9)

The apparatus of claim 9, wherein the first message further includes a PE tag comprising one of a node identifier and a PE identifier or a PE identifier.

### Claim 15 (depends on 9)

The apparatus of claim 9, further comprising circuitry and logic to:
send a second message comprising an RMO GET message from the initiator to the target, the RMO GET message including an RMO key, a destination buffer, and a length;
processing, via the NIC on the target, the RMO GET message to extract the RMO key and employ the RMO key to identify an address or address offset of a memory buffer in a memory;
read, via the NIC, data having the length beginning at the address or address offset in the memory buffer;
returning a third message from the target to the initiator including a data payload comprising the data that are read and the destination buffer; and
writing, at the initiator, the data payload into the destination buffer.

### Claim 16 (depends on 9)

The apparatus of claim 9, wherein the initiator has a plurality of processing elements (PEs) and one or more memories operatively coupled to a NIC, further comprising:
send a second message comprising an RMO GET message destined for a target from a network port, the RMO GET message including an RMO key, a source buffer on the target, and a length;
receive, at a network port, a third message from the target including the RMO key, a data payload comprising data that are read from the source buffer, and the length;
extract the RMO key and employ the RMO key to identify an address or address offset of a memory buffer in a memory; and
write the data payload at the address or address offset in the memory buffer.

### Claim 17

A compute node, configured to be implemented in a distributed environment comprising a plurality of compute nodes, comprising:
a plurality of processing elements (PEs);
one or more memory devices; and
a Network Interface Controller (NIC), including,
one or more network ports; and
circuitry and logic to:
receive, at a network port and originating from a second compute node comprising an initiator, a first message comprising a receiver-managed offset (RMO) PUT or UPDATE and including an RMO key, data, and a length;
extract the RMO key from the first message and employ the RMO key to identify an address or address offset of a memory buffer in a memory device; and
one of,
write the data to the memory buffer beginning at the address or address offset in the memory buffer when the first message is an RMO PUT; otherwise
apply an update operation to existing data at the address or address offset in the memory buffer using the data in an RMO UPDATE.

### Claim 18 (depends on 17)

The compute node of claim 17, wherein the NIC further comprises circuitry and logic to:
implement a list or table comprising a plurality of indicators indexed by an index value;
one of,
i. compute an index value using the RMO key as an input to a key function; or
ii. employ the RMO key as an index value;
look up an indicator in the list or table using the index value; and
employ the indicator to determine the address or address offset in the memory buffer.

### Claim 19 (depends on 17)

The compute node of claim 17, wherein the RMO key comprises a PE number comprising a first integer and an RMO number comprising a second integer, and where the key function comprises a bitwise concatenation of the PE number and the RMO number or a bitwise concatenation of the RMO number and PE number.

### Claim 20 (depends on 17)

The compute node of claim 17, wherein the compute node is further configured to:
send a second message comprising an RMO GET message destined for a target compute node from a network port, the RMO GET message including an RMO key, a source buffer on the target compute node, and a length;
receive, at a network port, a third message from the target compute node including the RMO key, a data payload comprising data that are read from the source buffer, and the length;
extract the RMO key from the third message and employ the RMO key to identify an address or address offset of a memory buffer in a memory device; and
write the data payload at the address or address offset in the memory buffer.
