REMOTE TRANSACTIONAL MEMORY

• US 20150269116A1
• Filed: 03/23/2015
• Published: 09/24/2015
• Est. Priority Date: 03/24/2014
• Status: Active Grant

• Current Assignee: Mellanox Technologies Limited (NVIDIA Corporation)
• Original Assignee: Mellanox Technologies Limited (NVIDIA Corporation)
• Inventors: Raikin, Shlomo, Liss, Liran, Shachar, Ariel, Bloch, Noam, Kagan, Michael
• Granted Patent: US 9,925,492 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: B01D 2251/2062 Ammonia
• CPC Class Codes: B01D 2255/20761 Copper
• CPC Class Codes: B01D 2255/50 Zeolites
• CPC Class Codes: B01D 2255/904 Multiple catalysts
• CPC Class Codes: B01D 2255/9155 Wall flow filters
• CPC Class Codes: B01D 53/9418 for removing nitrogen oxides by selective catalytic reduction [SCR] using a reducing agent in a lean exhaust gas
• CPC Class Codes: B01D 53/9477 with catalysts positioned on separate bricks, e.g. exhaust systems
• CPC Class Codes: B01J 2229/186 not in framework positions
• CPC Class Codes: B01J 29/072 Iron group metals or copper
• CPC Class Codes: B01J 29/56 Iron group metals or copper
• CPC Class Codes: B01J 29/76 Iron group metals or copper
• CPC Class Codes: B01J 29/763 CHA-type, e.g. Chabazite, LZ-218
• CPC Class Codes: B01J 35/19 Catalysts containing parts with different compositions
• CPC Class Codes: B01J 35/40 characterised by dimensions, e.g. grain size in a colloidal state B01J35/23; crystallite size B01J35/77
• CPC Class Codes: B01J 35/56 Foraminous structures having flow-through passages or channels, e.g. grids or three-dimensional monoliths
• CPC Class Codes: B01J 37/0215 Coating
• CPC Class Codes: B01J 37/0246 Coatings comprising a zeolite
• CPC Class Codes: B01J 37/08 Heat treatment B01J37/0009, B01J37/0018 take precedence
• CPC Class Codes: B01J 37/30 Ion-exchange
• CPC Class Codes: F01N 3/035 with catalytic reactors , e.g. catalysed diesel particulate filters
• CPC Class Codes: F01N 3/103 : Oxidation catalysts for HC and CO only
• CPC Class Codes: F01N 3/2066 : Selective catalytic reduction [SCR]
• CPC Class Codes: F01N 3/2842 : specially adapted for monolithic supports, e.g. of honeycomb type F01N3/2853 - F01N3/2871 take precedence
• CPC Class Codes: G06F 13/287 : Multiplexed DMA G06F13/30 takes precedence
• CPC Class Codes: G06F 15/17331 : Distributed shared memory [DSM], e.g. remote direct memory access [RDMA]
• CPC Class Codes: G06F 9/467 : Transactional memory G06F9/528 takes precedence
• CPC Class Codes: Y02C 20/10 : of nitrous oxide (N2O)

Abstract

Remote transactions using transactional memory are carried out over a data network between an initiator host and a remote target. The transaction comprises a plurality of input-output (IO) operations between an initiator network interface controller and a target network interface controller. The IO operations are controlled by the initiator network interface controller and the target network interface controller to cause the first process to perform accesses to the memory location atomically.

Claims

1. A method of communication over a data network, comprising the steps of:

performing a transaction across a data network between an initiator host and a remote target, the initiator host having an initiator network interface controller, and the remote target having a target network interface controller and a memory location that is accessible by a first process and a second process, the transaction comprising a plurality of input-output (IO) operations between the initiator network interface controller and the target network interface controller, respectively and

controlling the IO operations with the initiator network interface controller and the target network interface controller to cause the first process to perform accesses to the memory location atomically with respect to the second process.

2. The method according to claim 1, wherein the IO operations comprise messages including at least one of a first message to begin the transaction, a second message to commit the transaction and a third message that communicates a status of the transaction, further comprising responding to the messages with the target network interface controller.

3. The method according to claim 2, wherein responding to the first message comprises causes a state transition of the transaction from an idle state to an active state.

4. The method according to claim 2, wherein responding to the second message causes a state transition of the transaction from an active state to a committed state.

5. The method according to claim 2, wherein responding to the messages comprises in the target network interface controller associating a set of read and write operations in the memory location with the transaction.

6. The method according to claim 5, wherein associating a set of read and write operations comprises:

establishing a first table containing information on queue pairs of requester and responder processes that are participating in the transaction; and

establishing a second table containing transaction resource information as chunks of data.

7. The method according to claim 6, wherein a chunk includes a reference to the first table, an address of the chunk, an identifier of valid bytes for the chunk, a read/write indicator of whether the chunk has been read or written by currently active transactions;

a value to be committed in the transaction, and a pending-write indicator.

8. The method according to claim 2, wherein responding to the messages comprises generating a status message with the target network interface controller to report the status of the transaction to the initiator network interface controller.

9. The method according to claim 1, wherein contents of the memory location are available to the first process and concealed from the second process until an occurrence of a final state of the transaction.

10. The method according to claim 1, further comprising identifying accesses to the memory location with the target network interface controller that conflict with the transaction.

11. The method according to claim 10, further comprising the steps of:

responsively to identifying accesses transmitting an abort message to the initiator network interface controller with the target network interface controller to cause the transaction to abort; and

discarding results of store operations to the memory location that occurred during the transaction.

12. The method according to claim 1, wherein the IO operations are executed concurrently for a plurality of transactions.

13. The method according to claim 1, wherein the remote target comprises a plurality of remote targets and the IO operations occur between the initiator network interface controller and selected ones of the remote targets.

14. The method according to claim 13, wherein compute operations are performed responsively to the IO operations, and the compute operations are performed in the plurality of remote targets.

15. The method according to claim 1, further comprising transmitting command messages from the initiator network interface controller to the target network interface controller, wherein the accesses to the memory location are direct memory accesses that occur responsively to remote direct memory access requests in the command messages.

16. The method according to claim 1, wherein compute operations are performed responsively to the IO operations, and the compute operations are performed only in the remote target.

17. A network communication apparatus, comprising:

an initiator host interface, which is coupled to receive from an initiator host a request from an initiator process running on the initiator host to perform a transaction with a remote target via a data network, the remote target having a target network interface controller and a memory location that is accessible by a first process and a second process; and

a host network interface controller coupled to the initiator host and the data network, the transaction comprising a plurality of input-output (IO) operations between the host network interface controller and the target network interface controller, the host network interface controller being configured for controlling the IO operations by issuing commands to the target network interface controller to cause the first process to perform accesses to the memory location atomically with respect to the second process.

18. The apparatus according to claim 17, wherein the IO operations comprise messages including at least one of a first message to begin the transaction, a second message to commit the transaction and at least one third message that communicates a status of the transaction, wherein the host network interface controller is configured for responding to messages from the target network interface controller.

19. The apparatus according to claim 17, wherein contents of the memory location are available to the first process and concealed from the second process until an occurrence of a final state of the transaction.

20. The apparatus according to claim 17, wherein the host network interface controller is configured for performing the IO operations for a plurality of transactions concurrently.

21. The apparatus according to claim 17, wherein the remote target comprises a plurality of remote targets and the host network interface controller is configured for performing the IO operations with selected ones of the remote targets.

22. A network communication system, comprising:

a remote target having a memory location that is accessible by a first process and a second process;

a target network interface controller coupled to a data network;

an initiator host, and

a host network interface controller coupled to the initiator host and the data network, which is configured to receive from the initiator host a request from an initiator process running on the initiator host to perform a transaction with the remote target via the data network, the transaction comprising a plurality of input-output (IO) operations between the host network interface controller and the target network interface controller, wherein accesses to the memory location occur responsively to the IO operations, the host network interface controller and the target network interface controller being configured for conducting the IO operations to cause the first process to perform the accesses to the memory location atomically with respect to the second process, wherein the target network interface controller and a processor of the remote target are connected by a coherent bus.

23. The system according to claim 22, wherein the host network interface controller is configured for performing the IO operations for a plurality of transactions concurrently.