Patent Grant
10909072
The present disclosure relates generally to distributed computing environments, and more particularly, to a key value store snapshot in a distributed memory object architecture.
The computer industry continues to develop and refine solid-state storage devices and media, moving closer and closer to achieving memory-class storage. In the past decade there has been a thousand-fold reduction in access latency of affordable storage devices, and another twenty-fold reduction is expected within the year. At the same time, networking speeds have seen more than a 100-time increase in bandwidth with commensurate latency decrease, plus the emergence of standardized remote direct memory access (RDMA) functionality that can improve communication efficiency and further reduce latency.
These faster computing infrastructures demand new data infrastructures where both memory-speed data access and disk-like high storage density are strongly desired at the same time. Such new data infrastructures promise to bring significant performance improvements to computing tasks whose working data sets exceed dynamic random access memory (DRAM) capacity, and where highly frequent data movements between DRAM and lower storage tiers, such as solid state drive (SSD) and hard disk drive (HDD), are therefore required.
To provide the lowest possible access latency, operating system support of emerging persistent memory (PMEM) technology has created mechanisms for a user-space application to have direct access (DAX) to persistent memory media (i.e., without the access being performed by operating system software). Examples of existing solutions include:
“NOVA”, which is a single-node file system for persistent memory with emphasis on consistency. It uses per-file metadata journals for fast, concurrent, consistent updates. NOVA also supports DAX memory mapped access. It, however, does not provide cross-node replication or availability.
“Strata”, which is a single-node file system that provides a tiered, log-structured file system starting from a persistent memory layer and progressing to SSD then HDD as the data access frequency cools. It, however, does not support DAX memory map access, nor provide cross-node replication or availability.
“Octopus”, which is a multi-node distributed persistent memory file system using tightly integrated RDMA to reduce communication latency. It, however, does not support DAX memory mapped access.
“Hotpot”, which is a multi-node kernel-level distributed shared persistent memory system that provides low latency, transparent memory accesses, data persistence, data reliability, and high availability. It is focused on memory mapped access and does not address standard file storage IO operations.
“FluidMem”, which is a multi-node system that realizes disaggregated memory in the datacenter. It does not address memory persistence or storage IO.
None of these existing solutions, however, provide low-latency access of multi-node distributed data objects with both the semantics of memory and the semantics of file storage. It is therefore desirable to provide low-latency memory spaces: 1) that are accessible across a cluster of nodes, 2) that can exceed the memory capacity of a given node in the cluster, and 3) that can span the memory and storage of multiple nodes. It is further desirable that these memory spaces be accessible with either the load/store semantics of memory, or with the read/write, input/output semantics of file storage. Disclosed herein in a distributed memory object (DMO) system, referred to as MemVerge DMO system, that provides these types of low-latency memory spaces.
Disclosed herein is an apparatus and method for a key value store snapshot in a distributed memory object. In one embodiment, a method includes forming a system cluster comprising a plurality of nodes, wherein each node includes a memory, a processor and a network interface to send and receive messages and data; creating a plurality of sharable memory spaces having partitioned data, wherein each space is a distributed memory object having a compute node, wherein the sharable memory spaces are at least one of persistent memory or DRAM cache; storing data in persistent memory, the data having a generation tag created from a generation counter and a doubly linked list having a current view and a snapshot view, the data further being stored in either a root or a persisted row; creating a snapshot comprising a consistent point-in-time view of key value contents within a node and incrementing the generation counter; copying the snapshot to a second node; regenerating an index for the key value contents within the node; and logging updates since the snap was applied to update copied data in the second node.
In another embodiment, a key value store snapshot system for a distributed memory object comprises a system cluster comprising a plurality of nodes, wherein each node includes a memory, a processor and a network interface to send and receive messages and data, wherein the network interface operates on remote direct memory access; a plurality of sharable memory spaces having partitioned data, wherein each space is a distributed memory object having a compute node, wherein the sharable memory spaces are at least one of persistent memory or DRAM cache, wherein data is stored in persistent memory and the data has a generation tag created from a generation counter and a doubly linked list having a current view and a snapshot view, the data further being stored in either a root or a persisted row; and a snapshot comprising a consistent point-in-time view of key value contents within a node and incrementing the generation counter, wherein the snapshot is to be copied to a second node, and an index for the key value contents within the node is regenerated and updates are logged since the snap was applied to update copied data in the second node. Other embodiments are explained within this disclosure.
This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter. Furthermore, the claimed subject matter is not limited to implementations that solve any or all disadvantages noted in any part of this disclosure.
Various aspects of apparatuses and methods will now be presented in the detailed description by way of example, and not by way of limitation, with reference to the accompanying drawings, wherein:
As used in the following description, remote direct memory access (RDMA) refers to a direct memory access mechanism that enables a computer to access the memory of another computer without involving the operating system of either computer. Persistent memory (PMEM) refers to the storage of data structures such that the data can continue to be accessed using memory instructions, e.g., load and store, even after completion of the process that created or modified the data structures.
A MemVerge direct memory object (DMO) system provides persistent distributed memory objects that can be accessed as either in-memory or file-storage mode, and may be implemented in low-latency RDMA. Thus, the MemVerge DMO system enables use of DMOs both as memory and storage. The MemVerge DMO system also allows data in the system to be converted between in-memory and file-storage modes. In general, embodiment MemVerge DMO systems provide close-to-memory-speed data access which in turn can significantly relive data bottlenecks observed at upper layer applications. Furthermore, embodiments may be built in user space, thus obviating the need to install a customized operating system kernel. We now turn to
Address space for a DMO may be partitioned into equal size chunks, with each chunk being stored on one or more chunk replica (CR) nodes 144 included in the cluster of nodes 105. The chunks are distributed among a subset of the cluster nodes in such a manner as to: 1) focus locality of the chunks for performance efficiency, 2) provide sufficient availability of address space, and to 3) balance resources among the cluster of nodes. Furthermore, any node in a cluster using a DMO can locally keep a copy of a page.
The object owner node 142 is responsible for coordinating updates to the client proxy nodes 140 as well as the chunk replica nodes 144. The object owner node 142 is also responsible for maintaining a configurable replication factor per DMO. The object owner node 142 and chunk replica nodes 144 can migrate to deal with failures, performance, or resource constraints. Client proxy nodes 140 and chunk replica nodes 144 cooperate with the object owner node 142 in implementing protocols to make coherent updates and thereby provide a crash consistent view in the face of failures.
Various functional components of a DMO system are associated with one or more nodes in the system cluster and are described in the follow paragraphs.
Node Manager (NM)
A node manager (NM) 110 operates on each node in a MemVerge DMO system 100. Once a node manager 110 starts on a node, it can start or stop all other services associated with a node. Some services associated with a node may be started or stopped automatically or by request. The node manager 110 is responsible for finding or electing the cluster manager (CM) 114 and notifying its existence and node health to the cluster manager 114. Hence the node manager 110 has access to performance and exception information from other components in the system 100.
Cluster Manager (CM)
The cluster manager 114 runs on a single node in the MemVerge DMO system 100. The single node on which the cluster manager runs is elected by a consensus algorithm of the node managers. The cluster manager mediates cluster membership, node ID assignment, and the name service (NS) group 120. The cluster manager 114 also chooses nodes to satisfy allocation request constraints against cluster resource loading.
DMO Name Service (NS)
The DMO name service (NS) 124 is a hash-distributed service which provides mapping of a DMO name string to its object ID and the object owner. The service is hash distributed across a set of nodes in the system cluster 105. In the present example, the set of nodes is a name service group that is determined by the cluster manager.
Object Owner (OO)
The DMO object owner 142 is a single-node service that manages a DMO. The node corresponding to the client proxy 140 that creates the DMO becomes the object owner node 142. The object owner is responsible for selecting (via a cluster manager 114) an initial object node group 130 to contain the DMO and for assigning the chunk replicas (CRs) 144 within that node group. Some embodiments may contain additional object node groups 132, 134, etc. The object owner 142 also manages growing, shrinking, migrating, and recovering both the node group 130 as a whole, and the chunk replica 144 assignments within that group, as required to meet the DMO's size and replication requirement, or to optimize its usage efficiency. The object owner 142 can choose to move to another node (e.g., to be on the same node as a write client proxy). If the object owner 142 node fails, the DMO's node group will re-elect an object owner. The object owner keeps track of client proxies and orchestrates all updates affecting the DMO, e.g., configuration changes as well as data writes (msync commits and/or write TO).
Chunk Replica (CR)
The chunk replica 144 is a slave entity to the object owner 142 and client proxy 140. The object owner and client proxy read from and write to the chunk replica 144. The chunk replica owns some amount of storage devices (PMEM, SSD, etc.) on its node and manages the details of how/where a chunk of address space is stored therein.
Client Proxy (CP)
The client proxy 140 performs all input/output operations for the client and locally materializes and synchronizes/persists any object that the client requests to be memory mapped. To do that materialization, the client proxy creates a local cache for pieces of remote chunks that are in use and manages selection and eviction of pieces that are unused (or less actively used) as capacity constraints require. The client proxy 140 has code to specifically handle page fault notifications sent to it by the userfaultfd feature of Linux, or similar page fault notifications in other operating environments.
Example Operation Flows
Note that management of the cache capacity may require that a previously allocated area of cache be removed from its current role in the DMO address space (i.e., evicted) in order to reassign it for a new role. This eviction process can typically happen as a background task where an eviction candidate is selected, unmapped from the DMO space, and written back via an RDMA write to its remote location if required. The cache area of that candidate is then freed for reallocation.
With continued reference to
The client application starts using the DMO, i.e., it can do load/store references to the DMO, and/or read/write input/output calls to/from the DMO. If a load/store reference from the client application accesses a DMO region that is not over mapped, the client application takes/receives a page fault. The MemVerge library gets a page fault notification and calls to the client proxy. The client proxy caches the needed region into the cache file and replies to the MemVerge library. The MemVerge library then can over map the new region onto an appropriate local DMO space.
Thus, from a client application perspective, a MemVerge DMO system 100 enables a user, via the client application in conjunction with a client proxy, to initiate the use of a DMO, have data placed in one or more memory regions mapped to the DMO by either of a store call or a write call, and access data stored in one or more memory regions mapped to the DMO by a load call or a read call.
In general, high availability is an important feature for a storage system. This aspect is particularly important in a multi-node system when nodes are removed, and a failure happens and for load balancing when nodes are added. Embodiments disclosed herein leverage PMEM in these scenarios in a data rebuild process. That is, some embodiments are a PMEM based local KV storage that utilizes a snapshot that creates a consistent point-in-time view and can be used to migrate and rebuild KV data in another available node from a replica when an original node is removed. Snapshot embodiments are described below in more detail with reference to
For example, with reference to
With reference to
Implementation Alternatives
To implement larger memory space than physically available on a node, some form of demand paging is necessary. Three implementation approaches are presented: user space, kernel, and hypervisor. By way of example,
The data for the faulting area is copied to a local file and that file is mapped into the client's address space, thus materializing the data for the page fault. This mechanism is the least intrusive to the operating environment in that it does not require a new operating system. A user space approach may provide fast development, which in turn may be beneficial to both prototype and product developments. Furthermore, a user space approach may facilitate experiments with different new technologies invented for user space apps. Additionally, a user space approach may be easy to deploy, easy to containerize, and may easily connect to other user space storage frameworks such as SPDK. However, it may utilize indirect access to memory map.
In some embodiments, a kernel space approach uses the kernel memory management to intercept page faults from the client process. As a kernel entity, the page fault handler can directly manipulate the address map. The handler can maintain a memory pool where it caches a copy of the required data, and then directly map it into the client's address space. A kernel space approach can provide a fast, efficient access to memory map; however, it also could use a custom OS and may increase scope of testing (re-verify OS).
While various embodiments of the invention have been described above, they have been presented by way of example only, and not by way of limitation. Likewise, the various diagrams may depict an example architectural or other configuration for the disclosure, which is done to aid in understanding the features and functionality that can be included in the disclosure. The disclosure is not restricted to the illustrated example architectures or configurations, but can be implemented using a variety of alternative architectures and configurations. Additionally, although the disclosure is described above in terms of various exemplary embodiments and implementations, the various features and functionality described in one or more of the individual embodiments are not limited in their applicability to the particular embodiment with which they are described. They instead can be applied alone or in some combination, to one or more of the other embodiments of the disclosure, whether or not such embodiments are described, and if such features are presented as being a part of a described embodiment. Thus, the breadth and scope of the present disclosure should not be limited by any of the above-described exemplary embodiments.
In this document, the terms “module” and “engine” as used herein, refers to software, firmware, hardware, and any combination of these elements for performing the associated functions described herein. Additionally, for purpose of discussion, the various modules are described as discrete modules; however, as would be apparent to one of ordinary skill in the art, two or more modules may be combined to form a single module that performs the associated functions according embodiments of the invention.
In this document, the terms “computer program product”, “computer-readable medium”, and the like, may be used generally to refer to media such as, memory storage devices, or storage unit. These, and other forms of computer-readable media, may be involved in storing one or more instructions for use by processor to cause the processor to perform specified operations. Such instructions, generally referred to as “computer program code” (which may be grouped in the form of computer programs or other groupings), when executed, enable the computing system.
It will be appreciated that, for clarity purposes, the above description has described embodiments of the invention with reference to different functional units and processors. However, it will be apparent that any suitable distribution of functionality between different functional units, processors or domains may be used without detracting from the invention. For example, functionality illustrated to be performed by separate processors or controllers may be performed by the same processor or controller. Hence, references to specific functional units are only to be references to suitable means for providing the described functionality, rather than indicative of a strict logical or physical structure or organization.
Terms and phrases used in this document, and variations thereof, unless otherwise expressly stated, should be construed as open ended as opposed to limiting. As examples of the foregoing: the term “including” should be read as meaning “including, without limitation” or the like; the term “example” is used to provide exemplary instances of the item in discussion, not an exhaustive or limiting list thereof; and adjectives such as “conventional,” “traditional,” “normal,” “standard,” “known”, and terms of similar meaning, should not be construed as limiting the item described to a given time period, or to an item available as of a given time. But instead these terms should be read to encompass conventional, traditional, normal, or standard technologies that may be available, known now, or at any time in the future.
Likewise, a group of items linked with the conjunction “and” should not be read as requiring that each and every one of those items be present in the grouping, but rather should be read as “and/or” unless expressly stated otherwise. Similarly, a group of items linked with the conjunction “or” should not be read as requiring mutual exclusivity among that group, but rather should also be read as “and/or” unless expressly stated otherwise. Furthermore, although items, elements or components of the disclosure may be described or claimed in the singular, the plural is contemplated to be within the scope thereof unless limitation to the singular is explicitly stated.
The presence of broadening words and phrases such as “one or more,” “at least,” “but not limited to”, or other like phrases in some instances shall not be read to mean that the narrower case is intended or required in instances where such broadening phrases may be absent.
Additionally, memory or other storage, as well as communication components, may be employed in embodiments of the invention. It will be appreciated that, for clarity purposes, the above description has described embodiments of the invention with reference to different functional units and processors. However, it will be apparent that any suitable distribution of functionality between different functional units, processing logic elements or domains may be used without detracting from the invention. For example, functionality illustrated to be performed by separate processing logic elements or controllers may be performed by the same processing logic element or controller. Hence, references to specific functional units are only to be seen as references to suitable means for providing the described functionality, rather than indicative of a strict logical or physical structure or organization.
Furthermore, although individually listed, a plurality of means, elements or method steps may be implemented by, for example, a single unit or processing logic element. Additionally, although individual features may be included in different claims, these may possibly be advantageously combined. The inclusion in different claims does not imply that a combination of features is not feasible and/or advantageous. Also, the inclusion of a feature in one category of claims does not imply a limitation to this category, but rather the feature may be equally applicable to other claim categories, as appropriate.
This application claims the benefit of and priority to U.S. Provisional Patent Application Ser. No. 62/713,537, filed on Aug. 2, 2018 for “Distributed Memory Object Architecture that Enables Memory-Speed Data Access for both Memory Semantics and Storage Semantics in a Distributed Environment”, and 2) U.S. Non-provisional patent application Ser. No. 16/255,414, filed on Jan. 23, 2019 for “A Distributed Memory Object Architecture”, the entire disclosure of which are incorporated herein by references.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5787247 | Norin | Jul 1998 | A |
| 20060106832 | Ben-Dyke | May 2006 | A1 |
| 20130218840 | Smith | Aug 2013 | A1 |
| 20160004480 | Lakshman | Jan 2016 | A1 |
| 20170003899 | Raja | Jan 2017 | A1 |
| 20170220777 | Wang | Aug 2017 | A1 |
| Number | Date | Country | |
|---|---|---|---|
| 20200042496 A1 | Feb 2020 | US |
| Number | Date | Country | |
|---|---|---|---|
| 62713537 | Aug 2018 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 16255414 | Jan 2019 | US |
| Child | 16372283 | US |
