Technologies for transactional synchronization of distributed objects in a fabric architecture

• US 10,084,724 B2
• Filed: 09/09/2016
• Issued: 09/25/2018
• Est. Priority Date: 09/09/2016
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Bernat, Francesc Guim, Doshi, Kshitij A., Barragan, Daniel Rivas
• Primary Examiner(s): Wong, Warner
• Application Number:
  US15/260,626
• Publication Number:
  US 20180077086A1
• Time in Patent Office: 746 Days
• CPC Class Codes: G06F 12/084 with a shared cache
• CPC Class Codes: G06F 3/067 Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]
• CPC Class Codes: H04L 49/103 using a shared central buffer; using a shared memory
• CPC Class Codes: H04L 49/355 Application aware switches, e.g. for HTTP
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]

Abstract

Technologies for performing switch-based collective operations in a fabric architecture include a computing node that includes a host fabric interface (HFI) usable to communicate with network computing devices of the fabric architecture. The HFI is configured to associate an object with a transaction identifier generated by the HFI for a corresponding transactional synchronization session managed by the HFI of the computing node. Additionally, the HFI is configured to store received data associated with received write transaction that include the transaction identifier in a local buffer of the HFI. Upon receiving a transactional synchronization session termination request, the HFI is configured to initiate a write of the data stored in the local buffer of the HFI to one of the one or more data storage devices of the computing node. Other embodiments are described herein.

Claims

1. A computing node for performing switch-based collective operations in a fabric architecture in which the computing node is communicatively coupled to a plurality of interconnected computing nodes of the fabric architecture, the computing node comprising:

a processor that includes one or more processor cores;

a host fabric interface (HFI); and

one or more data storage devices having stored therein a plurality of instructions that, when executed by the processor, cause the computing node to;

generate, by the HFI and in response to having received a transactional synchronization session initialization request from an application presently executing on a processor core of the one or more processor cores, a transaction identifier for a transactional synchronization session;

transmit, by the HFI, the generated transaction identifier to the processor core;

associate, by the HFI and in response to having received an object association request that includes an identifier of an object and an access mode from the processor, the transaction identifier with the object identifier and the access mode of the object association request, wherein the access mode of the object association request is usable to determine whether the object association request comprises one of a write transaction request or a read transaction request;

determine, by the HFI and in response to a determination that the object association request corresponds to a write transaction request, whether the object is locally homed relative to the computing node, wherein the write transaction request includes data to be stored at a memory address associated with the object;

store, by the HFI and in response to a determination that the object is locally homed, the received data in a local buffer of the HFI; and

initiate, by the HFI and in response to having received a transactional synchronization session exit request from the processor core, a write of the data stored in the local buffer of the HFI to one of the one or more data storage devices.

2. The computing node of claim 1, wherein to receive a transactional synchronization session initialization request comprises to receive an instruction from the processor core via a software stack of the computing node, and wherein the instruction includes the session initialization request.

3. The computing node of claim 1, wherein to associate the transaction identifier with the object identifier and the access mode of the object association request comprises to (i) determine whether the object associated with the object identifier is locally homed relative to the computing node and (ii) store the transaction identifier, the object identifier, and the access mode in an entry of a monitoring table, wherein the monitoring table is usable to monitor transactions associated with the object during the transactional synchronization session.

4. The computing node of claim 1, wherein the object identifier comprises an address of the object in data storage device of the computing node.

5. The computing node of claim 1, wherein to store the received data in the local buffer of the HFI further comprises to store a local buffer address in a transaction storage table, wherein the local buffer address corresponds to a location of the stored data in the local buffer of the HFI.

6. The computing node of claim 1, wherein the plurality of instructions further cause the computing node to:

transmit, by the HFI and in response to having received a read transaction associated with the object from the processor core, a data retrieval request to a memory access manager of the computing node, wherein the data retrieval request is usable to retrieve data associated with the read transaction;

receive, by the HFI, the requested data from the memory access manager; and

forward, by the HFI, the received requested data to the processor core.

7. The computing node of claim 1, wherein the plurality of instructions further cause the computing node to:

detect, by the HFI, an object access violation;

broadcast, by the HFI, a software stack violation notification; and

delete, by the HFI, all object references associated with the object.

8. The computing node of claim 7, wherein to broadcast the software stack violation notification comprises to issue a local software stack interrupt.

9. The computing node of claim 7, wherein to delete all object references associated with the object comprises to delete one or more of (i) data associated with corresponding write transactions stored in a storage buffer of the HFI and (ii) object information corresponding to the object from a corresponding entry of a monitoring table usable to monitor transactions associated with the object during the transactional synchronization session.

10. One or more non-transitory, computer-readable storage media comprising a plurality of instructions stored thereon that in response to being executed cause a computing node to:

generate, by a host fabric interface (HFI) of the computing node and in response to having received a transactional synchronization session initialization request from an application presently executing on a processor core of a processor of the computing node, a transaction identifier for a transactional synchronization session;

transmit, by the HFI, the generated transaction identifier to the processor core;

associate, by the HFI and in response to having received an object association request that includes an identifier of an object and an access mode from the processor, the transaction identifier with the object identifier and the access mode of the object association request, wherein the access mode of the object association request is usable to determine whether the object association request comprises one of a write transaction request or a read transaction request;

determine, by the HFI and in response to a determination that the object association request corresponds to a write transaction request, whether the object is locally homed relative to the computing node, wherein the write transaction request includes data to be stored at a memory address associated with the object;

store, by the HFI and in response to a determination that the object is locally homed, the received data in a local buffer of the HFI; and

initiate, by the HFI and in response to having received a transactional synchronization session exit request from the processor core, a write of the data stored in the local buffer of the HFI to one of the one or more data storage devices.

11. The one or more non-transitory, computer-readable storage media of claim 10, wherein to receive a transactional synchronization session initialization request comprises to receive an instruction from the processor core via a software stack of the computing node, and wherein the instruction includes the session initialization request.

12. The one or more non-transitory, computer-readable storage media of claim 10, wherein to associate the transaction identifier with the object identifier and the access mode of the object association request comprises to (i) determine whether the object associated with the object identifier is locally homed relative to the computing node and (ii) store the transaction identifier, the object identifier, and the access mode in an entry of a monitoring table, wherein the monitoring table is usable to monitor transactions associated with the object during the transactional synchronization session.

13. The one or more non-transitory, computer-readable storage media of claim 10, wherein the object identifier comprises an address of the object in data storage device of the computing node.

14. The one or more non-transitory, computer-readable storage media of claim 10, wherein to store the received data in the local buffer of the HFI further comprises to store a local buffer address in a transaction storage table, wherein the local buffer address corresponds to a location of the stored data in the local buffer of the HFI.

15. The one or more non-transitory, computer-readable storage media of claim 10, wherein the plurality of instructions further cause the computing node to:

transmit, by the HFI and in response to having received a read transaction associated with the object from the processor core, a data retrieval request to a memory access manager of the computing node, wherein the data retrieval request is usable to retrieve data associated with the read transaction;

receive, by the HFI, the requested data from the memory access manager; and

forward, by the HFI, the received requested data to the processor core.

16. The one or more non-transitory, computer-readable storage media of claim 10, wherein the plurality of instructions further cause the computing node to:

detect, by the HFI, an object access violation;

broadcast, by the HFI, a software stack violation notification; and

delete, by the HFI, all object references associated with the object.

17. The one or more non-transitory, computer-readable storage media of claim 16, wherein to broadcast the software stack violation notification comprises to issue a local software stack interrupt.

18. The one or more non-transitory, computer-readable storage media of claim 16, wherein to delete all object references associated with the object comprises to delete at least one of (i) data associated with corresponding write transactions stored in a storage buffer of the HFI and (ii) object information corresponding to the object from a corresponding entry of a monitoring table usable to monitor transactions associated with the object during the transactional synchronization session.

19. A computing node for performing switch-based collective operations in a fabric architecture in which the computing node is communicatively coupled to a plurality of interconnected computing nodes of the fabric architecture, the computing node comprising:

means for generating, in response to having received a transactional synchronization session initialization request from an application presently executing on a processor core of a processor of the computing node, a transaction identifier for a transactional synchronization session;

means for transmitting the generated transaction identifier to the processor core;

means for associating, in response to having received an object association request that includes an identifier of an object and an access mode from the processor, the transaction identifier with the object identifier and the access mode of the object association request, wherein the access mode of the object association request is usable to determine whether the object association request comprises one of a write transaction request or a read transaction request;

means for determining, in response to a determination that the object association request corresponds to a write transaction request, whether the object is locally homed relative to the computing node, wherein the write transaction request includes data to be stored at a memory address associated with the object;

means for storing, in response to a determination that the object is locally homed, the received data in a local buffer of the HFI; and

means for initiating, in response to having received a transactional synchronization session exit request from the processor core, a write of the data stored in the local buffer of the HFI to one of the one or more data storage devices.

20. The computing node of claim 19, wherein the means for receiving a transactional synchronization session initialization request comprises means for receiving an instruction from the processor core via a software stack of the computing node, and wherein the instruction includes the session initialization request.

21. The computing node of claim 19, wherein the means for associating the transaction identifier with the object identifier and the access mode of the object association request comprises means for (i) determining whether the object associated with the object identifier is locally homed relative to the computing node and (ii) storing the transaction identifier, the object identifier, and the access mode in an entry of a monitoring table, wherein the monitoring table is usable to monitor transactions associated with the object during the transactional synchronization session.

22. The computing node of claim 19, wherein the means for storing the received data in the local buffer of the HFI further comprises means for storing a local buffer address in a transaction storage table, wherein the local buffer address corresponds to a location of the stored data in the local buffer of the HFI.

23. The computing node of claim 19, further comprising:

means for transmitting, in response to having received a read transaction associated with the object from the processor core, a data retrieval request to a memory access manager of the computing node, wherein the data retrieval request is usable to retrieve data associated with the read transaction;

means for receiving the requested data from the memory access manager; and

means for forwarding the received requested data to the processor core.

24. The computing node of claim 19, further comprising:

means for detecting an object access violation;

means for broadcasting a software stack violation notification; and

means for deleting all object references associated with the object.

25. The computing node of claim 24, wherein the means for deleting all object references associated with the object comprises means for deleting at least one of (i) data associated with corresponding write transactions stored in a storage buffer of the HFI and (ii) object information corresponding to the object from a corresponding entry of a monitoring table usable to monitor transactions associated with the object during the transactional synchronization session.