System and method for facilitating efficient management of non-idempotent operations in a network interface controller (NIC)

Abstract

A network interface controller (NIC) capable of efficient management of non-idempotent operations is provided. The NIC can be equipped with a network interface, storage management logic block, and an operation management logic block. During operation, the network interface can receive a request for an operation from a remote device. The storage management logic block can store, in a local data structure, outcome of operations executed by the NIC. The operation management logic block can determine whether the NIC has previously executed the operation. If the NIC has previously executed the operation, the operation management logic block can obtain an outcome of the operation from the data structure and generate a response comprising the obtained outcome for responding to the request.

Description

BACKGROUND

Field

This is generally related to the technical field of networking. More specifically, this disclosure is related to systems and methods for facilitating a network interface controller (NIC) with efficient management of non-idempotent operations.

Related Art

As network-enabled devices and applications become progressively more ubiquitous, various types of traffic as well as the ever-increasing network load continue to demand more performance from the underlying network architecture. For example, applications such as high-performance computing (HPC), media streaming, and Internet of Things (IOT) can generate different types of traffic with distinctive characteristics. As a result, in addition to conventional network performance metrics such as bandwidth and delay, network architects continue to face challenges such as scalability, versatility, and efficiency.

SUMMARY

A network interface controller (NIC) capable of efficient management of non-idempotent operations is provided. The NIC that can be equipped with a network interface, storage management logic block, and an operation management logic block. During operation, the network interface can receive a request for an operation from a remote device. The storage management logic block can store outcomes of operation executed by the NIC in a local data structure. The operation management logic block can determine whether the NIC has previously executed the operation. If the NIC has previously executed the operation, the operation management logic block can obtain an outcome of the operation from the data structure and generate a response comprising the obtained outcome for responding to the request.

BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 shows an exemplary network.

FIG. 2A shows an exemplary NIC chip with a plurality of NICs.

FIG. 2B shows an exemplary architecture of a NIC.

FIG. 3 shows an exemplary dynamic management process for non-idempotent operations in a NIC.

FIG. 4A shows a flow chart of a dynamic management process for non-idempotent operations in a source NIC.

FIG. 4B shows a flow chart of a dynamic management process for non-idempotent operations in a target NIC.

FIG. 5 shows an exemplary computer system equipped with a NIC that facilitates dynamic management of non-idempotent operations.

In the figures, like reference numerals refer to the same figure elements.

DETAILED DESCRIPTION

Various modifications to the disclosed embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the present disclosure. Thus, the present invention is not limited to the embodiments shown.

Overview

The present disclosure describes systems and methods that facilitate dynamic management for non-idempotent operations in a network interface controller (NIC). The NIC allows a host to communicate with a data-driven network.

The embodiments described herein solve the problem of ensuring proper execution of remotely-requested non-idempotent operations by (i) storing the outcome (or result) of the non-idempotent operations, which can be issued from a remote source, at a target NIC until acknowledged from the source, and (ii) upon receiving a request for an already-executed operation from the source, generating a response with the stored outcome. In this way, the NIC may avoid re-execution of a non-idempotent operation even if the source reissues the request.

During operation, an application, which may run on a source device of a NIC, can request a data operation (e.g., a “GET” or a “PUT” command of remote direct memory access (RDMA)) for a memory location of a remote target device. The NIC of the target device can receive the request, facilitate the execution of the operation, and send a response with the outcome of the execution. Examples of an outcome can include, but are not limited to, one or more values generated from a computation associated with the operation, an indicator that indicates successful or unsuccessful execution of the operation, a memory location or index associated with the operation, and information indicating the state of a data structure based on the execution of the operation on the data structure. The NICs of the source and target devices can be referred to as the source and target NICs, respectively. The operation can be an idempotent or a non-idempotent operation. An idempotent operation may be executed more than once without causing an error. On the other hand, a non-idempotent operation can be executed once. Executing a non-idempotent operation more than once may cause an error.

Typically, RDMA operations may include more non-idempotent operations than idempotent operations. Examples of non-idempotent operations can include, but are not limited to atomic memory operations, adding data to a queue, and indicating the completion of an operation. Since RDMA operations can be issued from a source device and transferred across a switch fabric to a target device, these operations are executed based on a request-response communication protocol between the source and target NICs. Under such circumstances, if a source NIC issues a request for an operation and does not receive a response, either the request or the response can be lost (e.g., due to packet loss in the switch fabric). For non-idempotent operations, such a loss may cause the source NIC to determine that operation has not been performed.

However, the source NIC may not distinguish whether the request or the response is lost. As a result, the source NIC may not be able to determine whether the target has already executed the operation. Consequently, the source NIC may reissue the request for the operation. Upon receiving the reissued request, the target NIC may re-execute the operation (e.g., if the response has been lost). Repeating the execution of a non-idempotent operation may cause an error. For example, if the operation is an enqueue operation, repeating the execution can cause the target NIC to erroneously enter an additional element in a queue.

To solve this problem, the source NIC can generate an indicator that indicates whether a requested operation is a non-idempotent operation. The source NIC can then include the request in a packet and send the packet to a target NIC via a switch fabric. The source NIC can include the indicator in the request (e.g., as a parameter of the request) or the packet (e.g., in the header of the packet). Upon receiving the request, the target NIC can determine that the requested operation is a non-idempotent operation based on the indicator. The target NIC can then store an indicator, which can be the received indicator or a locally generated one, indicating that the request is for a non-idempotent operation. The target NIC can also store the outcome of the operation in a local data structure. The target NIC can then generate a response based on the outcome, include the response in a packet, and send the packet to the source NIC.

If the target NIC receives a request for the same operation, the target NIC can determine that the request has been repeated by inspecting the stored outcomes. Accordingly, the target NIC can determine that the response sent to the source NIC may have been lost. Consequently, the target NIC can obtain the stored outcome from the data structure and return a response with the stored outcome, thereby avoiding the repetition of a non-idempotent operation. When the source NIC receives the response, the source NIC can send an acknowledgment piggybacked in a subsequent request message. Upon receiving the acknowledgment, the target NIC may remove the outcome from the data structure. The operation can be part of a sequence of operations. The source NIC can then use cumulative acknowledgment for a set of sequential responses. The target NIC may clear a corresponding set of outcomes based on the cumulative acknowledgment.

One embodiment of the present invention provides a NIC that can be equipped with a network interface, storage management logic block, and an operation management logic block. During operation, the network interface can receive a request for an operation from a remote device. The storage management logic block can store, in a local data structure, outcome of operations executed by the NIC. The operation management logic block can determine whether the NIC has previously executed the operation. If the NIC has previously executed the operation, the operation management logic block can obtain an outcome of the operation from the data structure and generate a response comprising the obtained outcome for responding to the request.

In a variation on this embodiment, the operation management logic block can determine whether the operation has been previously executed based on a sequence number of the operation and an expected sequence number.

In a variation on this embodiment, the operation management logic block can determine whether the operation is a non-idempotent operation based on an indicator in the request.

In a variation on this embodiment, if the operation has previously not been executed, the operation management logic block can execute the operation to generate the outcome.

In a further variation, the storage management logic block can store the generated outcome in the data structure.

In a variation on this embodiment, the storage management logic block may identify an acknowledgment associated with a sequence number in the request and remove a set of outcomes from the data structure based on the acknowledgment.

In a variation on this embodiment, the operation is in a set of sequential operations, which should be executed in order.

In a variation on this embodiment, the operation corresponds to an RDMA command.

In this disclosure, the description in conjunction with FIG. 1 is associated with the network architecture and the description in conjunction with FIG. 2A and onward provide more details on the architecture and operations associated with a NIC that supports efficient management of non-idempotent operations.

FIG. 1 shows an exemplary network. In this example, a network 100 of switches, which can also be referred to as a “switch fabric,” can include switches 102, 104, 106, 108, and 110. Each switch can have a unique address or ID within switch fabric 100. Various types of devices and networks can be coupled to a switch fabric. For example, a storage array 112 can be coupled to switch fabric 100 via switch 110; an InfiniBand (IB) based HPC network 114 can be coupled to switch fabric 100 via switch 108; a number of end hosts, such as host 116, can be coupled to switch fabric 100 via switch 104; and an IP/Ethernet network 118 can be coupled to switch fabric 100 via switch 102. In general, a switch can have edge ports and fabric ports. An edge port can couple to a device that is external to the fabric. A fabric port can couple to another switch within the fabric via a fabric link. Typically, traffic can be injected into switch fabric 100 via an ingress port of an edge switch, and leave switch fabric 100 via an egress port of another (or the same) edge switch. An ingress link can couple a NIC of an edge device (for example, an HPC end host) to an ingress edge port of an edge switch. Switch fabric 100 can then transport the traffic to an egress edge switch, which in turn can deliver the traffic to a destination edge device via another NIC.

Exemplary NIC Architecture

FIG. 2A shows an exemplary NIC chip with a plurality of NICs. With reference to the example in FIG. 1, a NIC chip 200 can be a custom application-specific integrated circuit (ASIC) designed for host 116 to work with switch fabric 100. In this example, chip 200 can provide two independent NICs 202 and 204. A respective NIC of chip 200 can be equipped with a host interface (HI) (e.g., an interface for connecting to the host processor) and one High-speed Network Interface (HNI) for communicating with a link coupled to switch fabric 100 of FIG. 1. For example, NIC 202 can include an HI 210 and an HNI 220, and NIC 204 can include an HI 211 and an HNI 221.

In some embodiments, HI 210 can be a peripheral component interconnect (PCI) or a peripheral component interconnect express (PCIe) interface. HI 210 can be coupled to a host via a host connection 201, which can include N (e.g., N can be 16 in some chips) PCle Gen 4 lanes capable of operating at signaling rates up to 25 Gbps per lane. HNI 210 can facilitate a high-speed network connection 203, which can communicate with a link in switch fabric 100 of FIG. 1. HNI 210 can operate at aggregate rates of either 100 Gbps or 200 Gbps using M (e.g., M can be 4 in some chips) full-duplex serial lanes. Each of the M lanes can operate at 25 Gbps or 50 Gbps based on non-return-to-zero (NRZ) modulation or pulse amplitude modulation 4 (PAM4), respectively. HNI 220 can support the Institute of Electrical and Electronics Engineers (IEEE) 802.3 Ethernet-based protocols as well as an enhanced frame format that provides support for higher rates of small messages.

NIC 202 can support one or more of: point-to-point message passing based on Message Passing Interface (MPI), remote memory access (RMA) operations, offloading and progression of bulk data collective operations, and Ethernet packet processing. When the host issues an MPI message, NIC 202 can match the corresponding message type. Furthermore, NIC 202 can implement both eager protocol and rendezvous protocol for MPI, thereby offloading the corresponding operations from the host.

Furthermore, the RMA operations supported by NIC 202 can include PUT, GET, and Atomic Memory Operations (AMO). NIC 202 can provide reliable transport. For example, if NIC 202 is a source NIC, NIC 202 can provide a retry mechanism for idempotent operations. Furthermore, connection-based error detection and retry mechanism can be used for ordered operations that may manipulate a target state. The hardware of NIC 202 can maintain the state necessary for the retry mechanism. In this way, NIC 202 can remove the burden from the host (e.g., the software). The policy that dictates the retry mechanism can be specified by the host via the driver software, thereby ensuring flexibility in NIC 202.

Furthermore, NIC 202 can facilitate triggered operations, a general-purpose mechanism for offloading, and progression of dependent sequences of operations, such as bulk data collectives. NIC 202 can support an application programming interface (API) (e.g., libfabric API) that facilitates fabric communication services provided by switch fabric 100 of FIG. 1 to applications running on host 116. NIC 202 can also support a low-level network programming interface, such as Portals API. In addition, NIC 202 can provide efficient Ethernet packet processing, which can include efficient transmission if NIC 202 is a sender, flow steering if NIC 202 is a target, and checksum computation. Moreover, NIC 202 can support virtualization (e.g., using containers or virtual machines).

FIG. 2B shows an exemplary architecture of a NIC. In NIC 202, the port macro of HNI 220 can facilitate low-level Ethernet operations, such as physical coding sublayer (PCS) and media access control (MAC). In addition, NIC 202 can provide support for link layer retry (LLR). Incoming packets can be parsed by parser 228 and stored in buffer 229. Buffer 229 can be a PFC Buffer provisioned to buffer a threshold amount (e.g., one microsecond) of delay bandwidth. HNI 220 can also include control transmission unit 224 and control reception unit 226 for managing outgoing and incoming packets, respectively.

NIC 202 can include a Command Queue (CQ) unit 230. CQ unit 230 can be responsible for fetching and issuing host side commands. CQ unit 230 can include command queues 232 and schedulers 234. Command queues 232 can include two independent sets of queues for initiator commands (PUT, GET, etc.) and target commands (Append, Search, etc.), respectively. Command queues 232 can be implemented as circular buffers maintained in the memory of NIC 202. Applications running on the host can write to command queues 232 directly. Schedulers 234 can include two separate schedulers for initiator commands and target commands, respectively. The initiator commands are sorted into flow queues 236 based on a hash function. One of flow queues 236 can be allocated to a unique flow. Furthermore, CQ unit 230 can further include a triggered operations module (or logic block) 238, which is responsible for queuing and dispatching triggered commands.

Outbound transfer engine (OXE) 240 can pull commands from flow queues 236 in order to process them for dispatch. OXE 240 can include an address translation request unit (ATRU) 244 that can send address translation requests to address translation unit (ATU) 212. ATU 212 can provide virtual to physical address translation on behalf of different engines, such as OXE 240, inbound transfer engine (IXE) 250, and event engine (EE) 216. ATU 212 can maintain a large translation cache 214. ATU 212 can either perform translation itself or may use host-based address translation services (ATS). OXE 240 can also include message chopping unit (MCU) 246, which can fragment a large message into packets of sizes corresponding to a maximum transmission unit (MTU). MCU 246 can include a plurality of MCU modules. When an MCU module becomes available, the MCU module can obtain the next command from an assigned flow queue. The received data can be written into data buffer 242. The MCU module can then send the packet header, the corresponding traffic class, and the packet size to traffic shaper 248. Shaper 248 can determine which requests presented by MCU 246 can proceed to the network.

Subsequently, the selected packet can be sent to packet and connection tracking (PCT) 270. PCT 270 can store the packet in a queue 274. PCT 270 can also maintain state information for outbound commands and update the state information as responses are returned. PCT 270 can also maintain packet state information (e.g., allowing responses to be matched to requests), message state information (e.g., tracking the progress of multi-packet messages), initiator completion state information, and retry state information (e.g., maintaining the information required to retry a command if a request or response is lost). If a response is not returned within a threshold time, the corresponding command can be stored in retry buffer 272. PCT 270 can facilitate connection management for initiator and target commands based on source tables 276 and target tables 278, respectively. For example, PCT 270 can update its source tables 276 to track the necessary state for reliable delivery of the packet and message completion notification. PCT 270 can forward outgoing packets to HNI 220, which stores the packets in outbound queue 222.

NIC 202 can also include an IXE 250, which provides packet processing if NIC 202 is a target or a destination. IXE 250 can obtain the incoming packets from HNI 220. Parser 256 can parse the incoming packets and pass the corresponding packet information to a List Processing Engine (LPE) 264 or a Message State Table (MST) 266 for matching. LPE 264 can match incoming messages to buffers. LPE 264 can determine the buffer and start address to be used by each message. LPE 264 can also manage a pool of list entries 262 used to represent buffers and unexpected messages. MST 266 can store matching results and the information required to generate target side completion events. MST 266 can be used by unrestricted operations, including multi-packet PUT commands, and single-packet and multi-packet GET commands.

Subsequently, parser 256 can store the packets in packet buffer 254. IXE 250 can obtain the results of the matching for conflict checking. DMA write and AMO module 252 can then issue updates to the memory generated by write and AMO operations. If a packet includes a command that generates target side memory read operations (e.g., a GET response), the packet can be passed to the OXE 240. NIC 202 can also include an EE 216, which can receive requests to generate event notifications from other modules or units in NIC 202. An event notification can specify that either a fill event or a counting event is generated. EE 216 can manage event queues, located within host processor memory, to which it writes full events. EE 216 can forward counting events to CQ unit 230.

Dynamic Operations Management in NIC

FIG. 3 shows an exemplary dynamic management process for non-idempotent operations in a NIC. In this example, devices 302 and 304 can be coupled with each other via switch fabric 320. Devices 302 and 304 can be equipped with NICs 310 and 202, respectively. During operation, an application 304 running on device 302 can issue a request 312 for a data operation 350 (e.g., an RDMA operation) for a memory location of device 304. Since device 302 is the source and device 304 is the target of request 312, devices 302 and 304 can be referred to as source and target devices, respectively, for request 312. NIC 202 can receive the request, facilitate the execution of operation 350, and send a response 314 with the outcome of the execution of operation 350. Examples of the outcome can include, but are not limited to, one or more values generated from a computation associated with operation 350, an indicator that indicates successful or unsuccessful execution of operation 350, a memory location or index associated with operation 350, and information indicating the state of a data structure based on the execution of operation 350 on the data structure.

Suppose that operation 350 is a non-idempotent operation. Since request 312 can be issued from device 302 and transferred across switch fabric 320 to device 304, operation 350 is executed based on a request-response communication protocol between devices 302 and 304. Under such circumstances, if device 302 issues request 312 and does not receive response 314, either request 312 or response 314 can be lost (e.g., due to packet loss in switch fabric 320). Such a loss may cause device 302 to determine that operation has not been performed. Since operation 350 can be a non-idempotent operation, device 302 may determine that operation 350 should be performed.

Device 302 may not distinguish whether request 312 or response 314 is lost. As a result, device 302 may not be able to determine whether operation 350 has been executed by device 304. Hence, device 302 may reissue request 312 for operation 350 even if device 304 has already executed operation 350. Consequently, upon receiving reissued request 312, device 304 may re-execute operation 350. Repeating the execution of non-idempotent operation 350 may cause an error. For example, if operation 350 is an enqueue operation, re-execution of operation 350 can cause device 304 to erroneously enter an additional element in a queue.

To solve this problem, NIC 310 can generate an indicator that indicates whether operation 350 is a non-idempotent operation. NIC 310 can then include request 312 in a packet and send the packet to NIC 202 via switch fabric 320. NIC 310 can include the indicator in request 312 or the packet. Upon receiving request 312, NIC 202 can determine that operation 350 is a non-idempotent operation based on the indicator. NIC 202 can then store an indicator, which can be the received indicator or a locally generated one, indicating that request 312 is for a non-idempotent operation. NIC 202 can also store outcome (or result) 336 of operation 350 in a local data structure 330. Data structure 330 can be referred to as a target result store (TRS). TRS 330 can be in target tables 278 of PCT 270 of NIC 202.

NIC 202 can then generate response 314 based on outcome 336, include response 314 in a packet, and sends the packet to NIC 302. If NIC 202 receives request 312 again, NIC 202 can determine that request 312 has been repeated by inspecting a sequence number of operation 350 and an expected sequence number. For example, if operation 350 is repeated, the sequence number of operation 350 can be smaller than the expected sequence number (e.g., between the sequence number of the last acknowledged response and the sequence number of the next expected new request). Accordingly, NIC 202 can determine that response 314 sent to NIC 310 may have been lost. Consequently, NIC 202 can obtain outcome 336 from TRS 330 and return response 314 with outcome 336. In this way, NIC 202 can reissue response 314 without repeating non-idempotent operation 350.

In some embodiments, operation 350 can be part of a set of ordered operations that should be executed sequentially. NIC 310 can include a sequence number in request 312 indicating the order of operation 350. Furthermore, NIC 310 can send an acknowledgment piggybacked request 312 for operations that have been successfully responded back to NIC 310. For example, TRS 330 can store outcomes 332 and 334 of operations that have been executed prior to the execution of operation 350. Upon receiving outcomes 332 and 334 from NIC 202, NIC 310 can include an acknowledgment for outcome 334.

Since the operations should be executed in order, the acknowledgment can cumulatively acknowledge outcomes 332 and 334. In other words, the acknowledgment can instruct NIC 202 to clear all outcomes up to outcome 334 from TRS 330. In the same way, when NIC 310 receives response 314, NIC 312 can send an acknowledgment piggybacked in a subsequent request message. Upon receiving the acknowledgment, NIC 202 may remove outcome 336 from TRS 330.

FIG. 4A shows a flow chart of a dynamic management process for non-idempotent operations in a source NIC. During operation, the NIC may receive a request for an operation (e.g., from a flow queue or a retry buffer of the NIC) (operation 402) and check whether the operation is an idempotent operation (operation 404). If the operation is an idempotent operation, the NIC can include the request in a packet and forward the packet to a target NIC (operation 414). On the other hand, if the operation is a non-idempotent operation, the NIC can determine a sequence of responses that have been successfully received (operation 406). The NIC can then generate an acknowledgment corresponding to the sequence (operation 408) and an indicator indicating a non-idempotent operation (operation 410). Subsequently, the NIC can include the request, the acknowledgment, and the indicator in the packet and forward the packet to the target NIC (operation 412). It should be noted that the request can be allocated to an MCU module and selected for forwarding based on arbitration, as described in conjunction with FIG. 2B.

FIG. 4B shows a flow chart of a dynamic management process for non-idempotent operations in a target NIC. During operation, the NIC may obtain a request for an operation from a packet received from a remote NIC (operation 452) and check whether the operation is an idempotent operation (operation 454). If the operation is a non-idempotent operation, the NIC can obtain an acknowledgment from the packet and determine a corresponding sequence (operation 456). The NIC can then clear the TRS based on the sequence (operation 458). Subsequently, the NIC can check whether the operation is a repeat operation (e.g., by comparing a sequence number of the operation and an expected sequence number) (operation 460).

If the operation is an idempotent operation (operation 454) or the operation is not a repeat operation (operation 460), the NIC can execute the operation to obtain an outcome and store the outcome in the TRS (operation 462). On the other hand, if the operation is a repeat operation (e.g., the sequence number of the operation is smaller than the expected sequence number), the NIC can obtain the outcome from the TRS (operation 466). Upon obtaining the outcome (operation 462 or 466), the NIC can include the outcome in a packet and forward the packet to the remote NIC (operation 464).

Exemplary Computer System

FIG. 5 shows an exemplary computer system equipped with a NIC that facilitates dynamic management of non-idempotent operations. Computer system 550 includes a processor 552, a memory device 554, and a storage device 556. Memory device 554 can include a volatile memory device (e.g., a dual in-line memory module (DIMM)). Furthermore, computer system 550 can be coupled to a keyboard 562, a pointing device 564, and a display device 566. Storage device 556 can store an operating system 570. An application 572 can operate on operating system 570.

Computer system 550 can be equipped with a host interface coupling a NIC 520 that facilitates efficient data request management. NIC 520 can provide one or more HNIs to computer system 550. NIC 520 can be coupled to a switch 502 via one of the HNIs. NIC 520 can include a PCT logic block 530, as described in conjunction with FIGS. 2B and 3. If NIC 520 operates as a source NIC, PCT logic block 530 can include an acknowledgment (ACK) logic block 532 that can generate cumulative acknowledgments for responses received from a target NIC. If NIC 520 operates as a target NIC, PCT logic block 530 can include an operation logic block 534 that can execute an operation associated with a request from a source NIC.

PCT logic block 530 can also include an TRS logic block 536 that can store the outcome of the execution of an operation in a local data structure (e.g., in memory device 554). TRS logic block 536 can remove (or clear) outcomes from the data structure based on cumulative acknowledgments. Operation logic block 534 can also determine whether an outcome of an operation is already stored in the data structure. If the outcome is already in the data structure, operation logic block 534 may obtain the outcome from the data structure instead of executing the operation. Upon generating the outcome, NIC 520 can include the outcome in a packet and forward the packet to the source NIC.

In summary, the present disclosure describes a NIC that facilitates efficient management of non-idempotent operations. The NIC can be equipped with a network interface, storage management logic block, and an operation management logic block. During operation, the network interface can receive a request for an operation from a remote device. The storage management logic block can store, in a local data structure, outcome of operations executed by the NIC. The operation management logic block can determine whether the NIC has previously executed the operation. If the NIC has previously executed the operation, the operation management logic block can obtain an outcome of the operation from the data structure and generate a response comprising the obtained outcome for responding to the request.

The methods and processes described above can be performed by hardware logic blocks, modules, or apparatus. The hardware logic blocks, modules, logic blocks, or apparatus can include, but are not limited to, application-specific integrated circuit (ASIC) chips, field-programmable gate arrays (FPGAs), dedicated or shared processors that execute a piece of code at a particular time, and other programmable-logic devices now known or later developed. When the hardware logic blocks, modules, or apparatus are activated, they perform the methods and processes included within them.

The methods and processes described herein can also be embodied as code or data, which can be stored in a storage device or computer-readable storage medium. When a processor reads and executes the stored code or data, the processor can perform these methods and processes.

The foregoing descriptions of embodiments of the present invention have been presented for purposes of illustration and description only. They are not intended to be exhaustive or to limit the present invention to the forms disclosed. Accordingly, many modifications and variations will be apparent to practitioners skilled in the art. Additionally, the above disclosure is not intended to limit the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A network interface controller (NIC), comprising: a network interface to receive a request for an operation from a remote device, wherein the request includes an acknowledgment corresponding to a sequence of responses successfully received by the remote device;a storage management logic block to: store, in a local data structure, outcome of operations executed by the network interface controller;obtain, from the request, the acknowledgement corresponding to the sequence of successfully received responses; andremove, from the data structure, a set of outcomes based on the sequence of successfully received responses; andan operation management logic block to: determine whether the operation has been previously executed by the network interface controller;in response to determining that the operation has previously been executed, obtain an outcome of the operation from the data structure; andgenerate a response comprising the obtained outcome for responding to the request.

2. The network interface controller of claim 1, wherein the operation management logic block is further to determine whether the operation has been previously executed based on a sequence number of the operation and an expected sequence number.

3. The network interface controller of claim 1, wherein the operation management logic block is further to determine whether the operation is a non-idempotent operation based on an indicator in the request.

4. The network interface controller of claim 1, wherein, in response to determining that the operation has previously not been executed, the operation management logic block is further to execute the operation to generate the outcome.

5. The network interface controller of claim 4, wherein the storage management logic block is further to store the generated outcome in the data structure.

6. The network interface controller of claim 1wherein a respective response of the sequence of responses is associated with a respective sequence number.

7. The network interface controller of claim 1, wherein the operation is among a set of sequential operations that are to be executed in order.

8. The network interface controller of claim 1, wherein the operation corresponds to a remote direct memory access (RDMA) command.

9. A method, comprising: receiving, by a network interface controller (NIC), a request for an operation from a remote device, wherein the request includes an acknowledgment corresponding to a sequence of responses successfully received by the remote device;storing, in a local data structure of the NIC, outcome of operations executed by the NIC;obtaining, from the request, the acknowledgement corresponding to the sequence of successfully received responses;removing, from the data structure, a set of outcomes based on the sequence of successfully received responses;determining whether the operation has previously been executed by the NIC;in response to determining that the operation has previously been executed, obtaining an outcome of the operation from the data structure; andgenerating a response comprising the obtained outcome for responding to the request.

10. The method of claim 9, further comprising determining whether the operation has been previously executed based on a sequence number of the operation and an expected sequence number.

11. The method of claim 9, further comprising determining whether the operation is a non-idempotent operation based on an indicator in the request.

12. The method of claim 9, further comprising, in response to determining that the operation has previously not been executed, executing the operation to generate the outcome.

13. The method of claim 12, further comprising storing the generated outcome in the data structure.

14. The method of claim 9, wherein a respective response of the sequence of responses is associated with a respective sequence number.

15. The method of claim 9, wherein the operation is among a set of sequential operations that are to be executed in order.

16. The method of claim 9, wherein the operation is a remote direct memory access (RDMA) command.

17. A computer system, comprising: a processor;a memory device; anda host interface to couple the NIC;wherein the NIC comprises: a network interface to receive a request for an operation from a remote device, wherein the request includes an acknowledgment corresponding to a sequence of responses successfully received by the remote device;a storage management logic block, in a data structure in the NIC, to: store outcomes of operations executed by the NIC;obtain, from the request, the acknowledgement corresponding to the sequence of successfully received responses; andremove, from the data structure, a set of outcomes based on the sequence of successfully received responses; andan operation management logic block to: determine whether the operation has previously been executed by the NIC;in response to determining that the operation has previously been executed, obtain an outcome of the operation from the data structure; andgenerate a response comprising the obtained outcome for responding to the request.

18. The computer system of claim 17, wherein the operation management logic block is further to determine whether the operation is a non-idempotent operation based on an indicator in the request.

19. The computer system of claim 17, wherein, in response to determining that the operation has not been executed, the operation management logic block is further to execute the operation to generate the outcome and store the generated outcome in the data structure.

20. The computer system of claim 17, wherein a respective response of the sequence of responses is associated with a respective sequence number.