GROUP-BASED DATA REPLICATION IN MULTI-TENANT STORAGE SYSTEMS

• US 20190294582A1
• Filed: 03/26/2018
• Published: 09/26/2019
• Est. Priority Date: 03/26/2018
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Zhu, Yibo, Padhye, Jitendra D., Liu, Hongqiang, Raindel, Shachar, Kim, Daehyeok, Badam, Anirudh
• Granted Patent: US 10,642,779 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 15/167 using a common memory, e.g. mailbox
• CPC Class Codes: G06F 16/27 Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
• CPC Class Codes: G06F 3/061 Improving I/O performance
• CPC Class Codes: G06F 3/0619 in relation to data integrity, e.g. data losses, bit errors
• CPC Class Codes: G06F 3/065 Replication mechanisms
• 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 67/1095 Replication or mirroring of data, e.g. scheduling or transport for data synchronisation between network nodes
• 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

Distributed storage systems, devices, and associated methods of data replication are disclosed herein. In one embodiment, a server in a distributed storage system is configured to write, with an RDMA enabled NIC, a block of data from a memory of the server to a memory at another server via an RDMA network. Upon completion of writing the block of data to the another server, the server can also send metadata representing a memory location and a data size of the written block of data in the memory of the another server via the RDMA network. The sent metadata is to be written into a memory location containing data representing a memory descriptor that is a part of a data structure representing a pre-posted work request configured to write a copy of the block of data from the another server to an additional server via the RDMA network.

Claims

1. A method for data replication in a distributed storage system having a plurality of storage nodes interconnected by an remote direct memory access (“

RDMA”

) network, the storage nodes individually having a processor, a memory, and an RDMA enabled network interface card (“

RNIC”

) operatively coupled to one another, the method comprising;

writing, from a first RNIC at a first storage node, a block of data from a first memory at the first storage node to a second memory at a second storage node via a second RNIC interconnected to the first RNIC in the RDMA network;

sending, from the first RNIC to the second storage node, metadata representing a memory location and a data size of the written block of data in the second memory via the second RNIC;

performing, at the second storage node, modification of a memory descriptor in the second memory according to the metadata, the memory descriptor being a part of a data structure representing a pre-posted work request for writing a copy of the block of data to a third storage node; and

upon completion of modifying the memory descriptor, writing, from the second RNIC, a copy of the block of data to a third memory at the third storage node via a third RNIC interconnected to the second RNIC in the RDMA network, thereby achieving replication of the block of data in the distributed storage system without using the processors at the second and third storage nodes.

2. The method of claim 1 wherein writing the block of data includes:

receiving, from a client device, data representing an update to a data object stored in the distributed storage system;

storing, in the first memory, a copy of the received data as the block of data at the first storage node; and

executing, at the first RNIC, another data structure representing another work request for writing the block of data to the second storage node, the another work request having parameters representing the memory address and data size of the block of data to be written in the second memory of the second storage node.

3. The method of claim 1 wherein modifying the file descriptor includes:

receiving, at the second storage node, the sent metadata from the first RNIC;

identifying, at the second RNIC, the memory location containing data representing the memory descriptor; and

updating, at the second memory, the memory descriptor at the identified memory location with the received metadata from the first RNIC.

4. The method of claim 1 wherein writing the copy of the block of data to the third memory includes:

determining whether the modification of the memory descriptor is completed; and

in response to determining that the modification of the memory descriptor is completed, at the second RNIC, automatically triggering writing, from the second storage node, the copy of the block of data to the third memory without using the processor at the second storage node.

5. The method of claim 1 wherein writing the copy of the block of data to the third memory includes:

determining whether the modification of the memory descriptor is completed; and

in response to determining that the modification of the memory descriptor is completed, at the second RNIC, automatically triggering a conditional execution work request at the second storage node; and

upon execution of the conditional execution work request at the second RNIC, writing, from the second storage node, the copy of the block of data to the third memory without using the processor at the second storage node.

6. The method of claim 1 wherein writing the copy of the block of data to the third memory includes:

determining whether the modification of the memory descriptor is completed; and

in response to determining that the modification of the memory descriptor is completed, at the second RNIC, automatically triggering a conditional execution work request at the second storage node, the conditional execution work request being pre-posted at the second storage node as a trigger for executing the pre-posted work request for writing the copy of the block of data to the third storage node; and

upon execution of the conditional execution work request at the second RNIC, writing, from the second storage node, the copy of the block of data to the third memory without using the processor at the second storage node.

7. The method of claim 1, further comprising:

upon completion of modifying the memory descriptor, sending, from the second RNIC, another metadata representing another memory address and data size of the written copy of the block of data in the third memory, to the third storage node via the third RNIC.

8. The method of claim 1, further comprising:

at the third storage node,

determining whether the third storage node is a last storage node in a replication group that includes the first, second, and third storage nodes; and

in response to determining that the third storage node is a last storage node in a replication group, transmitting a notification representing an acknowledgement of receiving a copy of the block of data at the second and third storage node to the first storage node upon receiving the sent metadata from the second storage node.

9. The method of claim 8, further comprising:

at the first storage node, upon receiving the notification representing the acknowledgement from the third storage node, sending additional metadata to the second storage node for committing the block of data to a persistent storage in the second storage node, the additional metadata including data representing a replication group ID, a memory offset of a memory region and a destination memory region, and a size of the block of data being copied; and

at the second storage node, upon receiving the additional metadata,

updating another memory descriptor of another data structure representing another pre-posted work request for writing the copy of the block of data to the persistent storage at the second storage node; and

upon completion of updating the another memory descriptor, automatically triggering, by the second RNIC, the another pre-posted work request to write the copy of the block of data to the persistent storage at the second storage node.

10. A method for data replication in a distributed storage system having a plurality of storage nodes interconnected by an remote direct memory access (“

RDMA”

) network, the storage nodes individually having a processor, a memory, and an RDMA enabled network interface card (“

RNIC”

) operatively coupled to one another, the method comprising;

sending, from a first RNIC at a first storage node, metadata to a memory at a second storage node via a second RNIC interconnected to the first RNIC in the RDMA network, the metadata representing memory offsets of a source region and a destination region and a data size of data to be moved;

receiving, at the second storage node, the sent metadata from the first storage node;

performing, at the second storage node, modification of a memory descriptor in the memory according to the metadata, the memory descriptor being a part of a data structure representing a pre-posted work request for writing a block of data from a first memory region to a second memory region in the memory of the second storage node; and

upon completion of modifying the memory descriptor, automatically triggering writing, by the second RNIC, a block of data having the data size from the source region of the memory to the destination region of the memory at the second storage node according to the corresponding memory offsets included in the metadata without using the processor for the writing operation at the second storage node.

11. The method of claim 10 wherein performing the modification includes:

identifying, at the second RNIC, the memory location containing data representing the memory descriptor; and

updating, at the memory, the memory descriptor at the identified memory location with the received metadata from the first RNIC.

12. The method of claim 10 wherein performing the modification includes:

identifying, at the second RNIC, the memory location containing data representing the memory descriptor; and

replacing current data at the identified memory location with the received metadata from the first RNIC.

13. The method of claim 10 wherein:

the memory descriptor is a first memory descriptor; and

the method further includes performing, at the second storage node, another modification of a second memory descriptor in the memory of the second storage node according to the metadata, the second memory descriptor being a part of a data structure representing a pre-posted work request for sending a copy of the metadata to another memory at a further storage node via the RDMA network.

14. The method of claim 10 wherein:

the memory descriptor is a first memory descriptor; and

the method further includes

performing, at the second storage node, another modification of a second memory descriptor in the memory of the second storage node according to the metadata, the second memory descriptor being a part of a data structure representing a pre-posted work request for sending a copy of the metadata to another memory at a further storage node via the RDMA network; and

upon completion of writing the block of data from the source region to the destination region in the memory at the second storage node, automatically triggering, at the second RNIC, execution of the pre-posted work request for sending a copy of the metadata to another memory.

15. The method of claim 10 wherein:

the memory descriptor is a first memory descriptor of a pre-posted first work request in a first working queue in the memory of the second storage node; and

the method further includes performing, at the second storage node, another modification of a second memory descriptor in the memory of the second storage node according to the metadata, the second memory descriptor being a part of a data structure representing a pre-posted second work request in a second working queue at the second storage node, the second work request being configured to send a copy of the metadata to another memory at a further storage node via the RDMA network.

16. A server in a distributed storage system having a plurality of servers interconnected by an remote direct memory access (“

RDMA”

) network, the server comprising;

a processor;

a memory; and

an RDMA enabled network interface card (“

RNIC”

) operatively coupled to one another, wherein the memory containing instructions executable by the processor to cause the server to;

write, with the RNIC, a block of data from the memory to another memory at another server via another RNIC interconnected to the RNIC in the RDMA network, the block of data representing an update to a data object stored in the distributed storage system;

upon completion of writing the block of data to the another server, send, from the RNIC to the another server, metadata representing a memory location and a data size of the written block of data in the another memory of the another server via the RDMA network; and

wherein the sent metadata is written into a memory location containing data representing a memory descriptor that is a part of a data structure representing a pre-posted work request configured to write a copy of the block of data to a further server via the RDMA network.

17. The server of claim 16 wherein:

the memory of the server contains a working queue having a WRITE work request; and

wherein execution of the WRITE work request by the RNIC writes the block of data from the memory to the another memory at another server.

18. The server of claim 16 wherein:

the memory of the server contains a working queue having a WRITE work request and a SEND work request in sequence;

wherein execution of the WRITE work request by the RNIC writes the block of data from the memory to the another memory at another server; and

wherein execution of the SEND work request sends the metadata to the another server.

19. The server of claim 16 wherein:

the memory of the server contains a working queue having a SEND work request;

the another memory of the another server contains another working queue having a RECV work request corresponding to the SEND work request; and

wherein the SEND and RECV work requests are configured to cooperate to send the metadata to the another server.

20. The server of claim 16 wherein the memory contains additional instructions executable by the processor to cause the server to:

receive, from the further server, a notification of acknowledgement indicating that both the another and further servers have a copy of the block of data; and

in response to receiving the notification, executing, at the RNIC, another SEND work request instructing the another and further servers to move the corresponding copy of the block of data from a source region to a destination region in the corresponding memories.