KR20190096838A - Method and apparatus for sending barrier command using dummy io request - Google Patents

Abstract

Disclosed is a method for transferring a barrier command using a dummy IO request. According to one embodiment of the present invention, the method which allows a host to guarantee a recording order of data recorded in a storage comprises the steps of: calling at least one order guaranteeing write request for supporting the guarantee of the recording order in the storage for at least one piece of block data, which is data corresponding to an updated data block; calling a barrier-based synchronization command of recording the block data in the storage before subsequent block data requested for the order guaranteeing write after the calling of the order guaranteeing write request; and calling a barrier-based write request for supporting the guarantee of the recording order in the storage and breakpoint setting of the recording order for dummy data.

Description

Barrier command delivery method using dummy input / output request and device therefor {METHOD AND APPARATUS FOR SENDING BARRIER COMMAND USING DUMMY IO REQUEST}

The present invention relates to a method of delivering a barrier command for guaranteeing an order in which data is written to storage using a dummy input / output request, and a computing device to which the method is applied.

The storage receives a command from an operating system and performs an operation corresponding to the command. For example, to send a command to flash memory-based storage, the Linux kernel must allocate an I / O request, set a command to pass to the I / O request, and pass it to that storage. Thus, the only way the host can send commands to the storage is through these I / O commands. The exception is that a flush, a command for storing data in a cache in a storage device in a persistent storage device, may be delivered without a specific input / output request for the storage.

In other words, in order for an existing Linux kernel to pass non-flush commands to storage, it must read data from or write data to storage. If the host has no data to read or write, there is no way to send a command to the storage.

Here, a flush command can be passed to storage by a user calling a system call such as fsync () or fdatasync (), but a barrier command that sets a break in the recording order between before and after data is only available when there is data to be written. There is a problem that can be delivered to storage. If the host calls a barrier related system call to ensure data write order, if there is no data to write, the barrier command is not sent to the storage device and the order of writing of data is not guaranteed.

Accordingly, there is a need for a method of delivering a barrier command to storage even when there is no data to be written using a dummy input / output request, and a computing device to which the method is applied.

Related prior arts include Korean Laid-Open Patent Publication No. 10-2017-0141184 (name of the invention: a logging device and a logging method).

An object of the present invention is to provide a method for transmitting a barrier command to storage by using a dummy input / output request to a storage device and a computing device to which the method is applied even when there is no data to be written to the storage.

The problem to be solved by the present invention is not limited to the problem (s) mentioned above, and other object (s) not mentioned will be clearly understood by those skilled in the art from the following description.

A barrier command transfer method using a dummy input / output request according to an embodiment of the present invention for achieving the above object is to ensure the recording order in the storage for at least one block data that is data corresponding to the updated data block Invoking at least one order guaranteed write request to support; Invoking a barrier-based synchronization command which is a command to cause the at least one block data to be written to the storage prior to subsequent block data called after the at least one ordered write request is invoked; Invoking a barrier-based write request for guaranteeing the writing order in the storage and setting a break point of the writing order for the dummy data.

In addition, a barrier command transfer method using a dummy input / output request according to another embodiment of the present invention for achieving the above object, the at least one block data of the data corresponding to the updated data block, the order of the recording order in the storage; Invoking at least one order guarantee write request supporting the guarantee; Invoking a barrier-based synchronization command which is a command to cause the at least one block data to be written to the storage prior to subsequent block data called after the at least one ordered write request is invoked; Determining whether the at least one order-guaranteed write request remains in an IO scheduler queue used to construct a write command from a write request; And if the at least one write request remains, invoking a barrier-based write request on dummy data to support guaranteeing the write order in the storage and setting breakpoints in the write order.

Preferably, when all or part of the at least one write request remains, the method may further include changing a write request consisting of a last write command among the at least one write request to a barrier-based write request.

In addition, a computing device for transmitting a barrier command by using a dummy input / output request according to an embodiment of the present invention for achieving the above object includes a storage in which software for operating a file system to support the guarantee of the recording order; And a processor for operating software mounted in the storage, the processor for at least one block data that is data corresponding to an updated data block, at least one order for guaranteeing a recording order in the storage; A barrier-based synchronization command that invokes a guarantee write request and causes the at least one block data to be written to the storage prior to subsequent block data called after the at least one order guarantee write request is invoked. To determine whether the at least one order-guaranteed write request remains in an IO scheduler queue used to construct a write command from a write request, and if the at least one write request remains, the storage for dummy data. Of record order in Invokes a barrier-based write request that supports setting breakpoints in chapter and record order.

Advantageously, if all or part of the at least one write request remains, the processor may change the write request consisting of the last write command of the at least one write request into a barrier-based write request.

According to the present invention, the barrier command is transmitted to the storage using a dummy input / output request, so that the barrier command can be delivered to the storage even when there is no data to be written to the storage.

1 is a diagram illustrating a data input / output process between a host and a storage device according to the related art.
FIG. 2 is a diagram illustrating an operation process of fsync () which is a synchronization command of an EXT4 file system according to the related art.
3 is a diagram for describing an order-guaranteed input and output stack according to an embodiment of the present invention.
4 is a flowchart illustrating a barrier command transfer method using a dummy input / output request according to an embodiment of the present invention.
5 is a flowchart illustrating a barrier command transfer method using a dummy input / output request according to another embodiment of the present invention.
6 is a diagram illustrating a barrier command transfer method using a dummy input / output request according to another embodiment of the present invention.
7 is a block diagram illustrating a computing device for transferring a barrier command using a dummy input / output request according to an embodiment of the present invention.

As the invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention. In describing the drawings, similar reference numerals are used for similar elements.

Terms such as first, second, A, and B may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component. The term and / or includes a combination of a plurality of related items or any item of a plurality of related items.

When a component is referred to as being "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may be present in between. Should be. On the other hand, when a component is said to be "directly connected" or "directly connected" to another component, it should be understood that there is no other component in between.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

1 is a diagram illustrating a data input / output process between a host and a storage device according to the related art.

Referring to FIG. 1, a file system operated by a host inputs a write request to an IO scheduler queue. The block device driver then removes one or more write requests from the IO scheduler queue and constructs a write command therefrom. The block device driver then delivers the write command to the storage device when possible. At this time, if the command queue of the storage device is not full, this may be the case. In addition, the storage controller removes the transferred write command from the command queue, receives the data related to the write command from the host, and stores the data in the cache. When the reception is complete, the storage sends a signal to the host. On the other hand, data stored in the cache is written to the storage surface periodically or by explicit request from the host.

Here, when considering the input order (I), write command transfer order (D), data transfer order (X), and data write order (P) of the write request, there may be a discrepancy in the following order.

First, I and D may be inconsistent. This is because the IO scheduler can reorder or coalesce write requests according to scheduling principles.

Also, D and X may be inconsistent. This is because the storage controller can freely adjust the order of write commands in the command queue.

In addition, X and P may be inconsistent. This is because the cache of storage is not a FIFO. That is, in flash-based storage, the order of data writing may be determined according to the order of writing in the associated mapping table, not the order of writing to the cache.

Due to these various ordering uncertainties, the data input / output process using the modern IO stack has been recognized as difficult to keep in order.

FIG. 2 is a diagram illustrating an operation process of fsync () which is a synchronization command of an EXT4 file system according to the related art.

Extended file system 4 (EXT4) is Linux's leading journaling file system, an improved version of the existing EXT3 file system. The journaling file system is a file system that tracks and records changes made in the journal before committing them to the file system. Through this, it is possible to minimize the damage of the system that can be caused when a system crash or power problems occur.

The EXT4 file system uses the fsync () function to control the recording order. In ordered mode (default), the data block (D) is written prior to the journal transaction.

First, the file system enters a write request for the data block (D). In this case, the data block D may be a plurality of blocks for other files. The application thread then waits for the DMA transfer to complete. When the DMA transfer is complete, the application thread runs the JBD thread to reflect the journal transaction. After that, the application thread goes back to sleep, and after the JBD thread has recorded all the journal transactions, it returns.

More specifically, the journal transaction can be reflected through two write requests. These are journal descriptor blocks and log blocks (JD) and commit blocks (JC). Also, in the course of reflecting journal transactions, JBD threads need to maintain their write order within a single transaction and between transactions.

Inside a single transaction, the JBD thread needs to write the JD before the JC. In other words, JBD threads can use the Transfer-and-Flush technique so that JDs are written before JC. In this case, the Transfer-and-Flush technique waits while data about JD is transferred through DMA, is written to storage, and then data about JC is transferred through DMA, and the storage is performed. In order to be flushed, the order of writing is kept.

In addition, between transactions, the JBD thread needs to ensure that the transactions are recorded in the order of the journal transaction's request.

As described above, in the EXT4 file system, a plurality of context switches occur during synchronization using fsync (), and a waiting time for data transmission and writing is required, which inevitably causes transmission overhead and flush overhead.

3 is a diagram for describing an order-guaranteed input and output stack according to an embodiment of the present invention.

An ordered I / O stack according to an embodiment of the present invention is a barrier-based write command, an order preserving dispatch module, and an epoch-based IO scheduler newly defined in an ordered block device layer. Can be configured. By using such configurations, the order of the write requests can be partially reflected in the storage, without using the Transfer-and-Flush technique.

More specifically, the order guarantee block device layer divides the write request into two types. These are ordered write requests and unordered write requests. Here, the order guarantee write request is a write request for ensuring the storage write order.

Also, a set of order-guaranteed write requests, in which the units whose recording order can be changed are epochs. In addition, one special type of order guaranteed write commands corresponding to the order guaranteed write request is a barrier-based write command. That is, barrier-based write commands are used to classify epochs, and the write order in storage between the order-safe write commands before and after it is guaranteed.

In this case, new attributes REQ_ORDERED and REQ_BARRIER may be used for the order-safe write command and the barrier-based write command, respectively. That is, REQ_ORDERED may be used for a general order guarantee write command, and REQ_ORDERED and REQ_BARRIER may be used together for a barrier-based write command.

4 is a flowchart illustrating a barrier command transfer method using a dummy input / output request according to an embodiment of the present invention.

In step S410, the host calls at least one order guarantee write request for guaranteeing the write order in the storage for at least one block data which is data corresponding to the updated data block.

In this case, the write request of the host may be classified into an order guarantee write request capable of order guarantee and an unordered write request that does not guarantee order. If a host needs to guarantee the order of writing of data, it can call the order-safe write request to ensure that writing order.

For example, when a host creates or modifies a specific file, the host may call at least one order-guaranteed write request for at least one block data for the generated or modified portion of the file.

In addition, the at least one ordered write request that is called is input to the IO scheduler queue, and may be configured by a different process and composed of a plurality of ordered write commands and transferred to the storage. However, data corresponding to a plurality of order-guaranteed write commands delivered to the storage is not yet permanently written to the flash memory included in the storage, but may be stored in the cache memory of the storage.

In step S420, the host invokes a barrier-based synchronization command, which is a command that causes the at least one block data to be written to the storage before the subsequent block data called after the at least one order guarantee write request is called. .

That is, the host may call at least one order guarantee write request for at least one block data before the order guarantee write request for subsequent block data so that the at least one block data is written to the storage before the subsequent block data. Can be. To this end, the host may call a barrier-based synchronization command so that at least one block data is preferentially written to storage.

For example, after the host has written the A file to storage, the host invokes at least one order-guaranteed write request write (A) to write the B file to the storage, followed by a barrier-based synchronization command fbarrier (A). Can be called. In addition, the host may call at least one order guarantee write request (write (B)) so that the A file is written to the storage before the B file.

Finally, in step S430, the host invokes a barrier-based write request for supporting dummy data and guaranteeing a writing order in storage and setting a break point of the writing order.

That is, the host may call a barrier-based write request for dummy data that is not substantially written to a file, thereby ensuring that the data requested for writing before and after the barrier-based write request is written on storage. In this case, the host may call the barrier-based write request regardless of whether at least one order-guaranteed write request remains in the IO scheduler queue.

As a result, previously written request data of the barrier based write request forms a single epoch. In addition, the host cannot guarantee the recording order between a plurality of order-guaranteed write requests included in a single epoch through a barrier-based write request. However, the host completes the formation of the epoch while guaranteeing the recording order between epochs, and thus is a break point of the write order. Can be set.

To this end, the host can change the invoked barrier-based write request into an order-safe write request. After blocking the write request that is input to the IO scheduler queue, the at least one ordered write request included in the IO scheduler queue is transmitted to the dispatch queue, and the last ordered write request is changed to a barrier-based write request. Can transmit Then, the input blocking to the IO scheduler queue can be released.

For example, the host may call a barrier-based write request for dummy data by designating a memory address and a size of data to be stored to a predetermined value (eg, 0). This allows the host to guarantee the recording order of the data without actually storing the data in a file.

As described above, the barrier command transfer method using the dummy input / output request according to an embodiment of the present invention transmits a barrier command corresponding to the barrier-based write request to the storage by using the dummy input / output request, so that there is no data to be written to the storage. In this case, there is an effect of passing a barrier command to storage.

5 is a flowchart illustrating a barrier command transfer method using a dummy input / output request according to another embodiment of the present invention.

In step S510, the host calls at least one order guarantee write request that supports the guarantee of the recording order in the storage for at least one block data which is data corresponding to the updated data block.

In step S520, the host calls a barrier-based synchronization command, which is a command that causes the at least one block data to be written to the storage before the subsequent block data called after the at least one order guarantee write request is called.

In step S530, the host determines whether at least one order-guaranteed write request remains in the IO scheduler queue used to construct a write command from the write request.

For example, the at least one ordered write request stored in the IO scheduler queue by another process running on the processor of the host may be composed of at least one ordered write command and transmitted to the storage.

As a result, all or some of the write requests of at least one ordered write request stored in the IO scheduler queue may remain, or conversely, none of the at least one ordered write request may remain.

Finally, in step S540, if the host does not have at least one write request left, the host invokes a barrier-based write request that supports guaranteeing the write order in storage and setting a break point of the write order for the dummy data.

For example, if at least one ordered write command corresponding to the at least one ordered write request is delivered to the storage by another process running on the processor of the host, the at least one write request remains in the IO scheduler queue. You may not.

In this case, the host may call a barrier-based write request for dummy data by designating a memory address and a size of data to be stored as a predetermined value (eg, 0).

In another embodiment, the host may change the write request consisting of the last write command of the at least one write request to a barrier-based write request if all or part of the at least one write request remains.

If all or some of the at least one write request remain in the IO scheduler queue, the host changes the write request consisting of the last write command among the at least one write request to a barrier-based write request (S550). You can configure base write commands and send them to storage.

That is, the host may call a barrier-based write request for dummy data or change one of the existing order-guaranteed write requests to a barrier-based write request, depending on whether the at least one write request remains in the IO scheduler queue.

Meanwhile, referring to FIG. 6, a barrier command transfer method using a dummy input / output request according to another embodiment of the present invention will be described.

That is, Figure 6 shows a barrier-based synchronization command (e.g. fbarrier ()) between each write to ensure the order of the data written to each file when the host's process writes data to file A and file B, respectively. , But the data in file A is already written by another process, indicating that there is no data to be written by the barrier-based synchronization command.

More specifically, first, the process P1 of the host writes data to file A (①). At this time, a context switch occurs to another process P2, and the other process P2 writes the data of the file A to the storage device (2). The context switch back to the original process P1, and the original process P1 invokes a barrier based synchronization command to ensure that the data in file A is first written to storage before writing the data to file B ( ③). Process P1 confirms that there is no data in file A to be written to storage, and allocates a dummy input / output request (④). After setting the barrier command in the dummy I / O request (⑤), it delivers it to the storage (⑥). Finally, write data to file B (⑦).

7 is a block diagram illustrating a computing device for transferring a barrier command using a dummy input / output request according to an embodiment of the present invention.

Referring to FIG. 7, a computing device 700 that transmits a barrier command using a dummy input / output request according to an embodiment of the present invention includes a storage 710 and a processor 720.

The storage 710 is equipped with software for operating a file system that supports the guarantee of the recording order.

The processor 720 runs software mounted on the storage 710.

In this case, the processor 720 calls at least one order-guaranteed write request that supports the guarantee of the write order in the storage 710 with respect to at least one block data that is data corresponding to the updated data block, and at least the Invoke a barrier-based synchronization command, which is a command that causes one block data to be written to storage 710 before the subsequent block data called after the at least one ordered write request is invoked, and from the write request. Determine if there is at least one order-guaranteed write request left in the IO scheduler queue used to construct the write command, and if the at least one write request is not left, the write order in the storage 710 for the dummy data. Invokes a barrier-based write request that supports the establishment of breakpoints in guarantee and write order.

In another embodiment, if all or part of the at least one write request remains, the processor 720 may change the write request consisting of the last write command among the at least one write request into a barrier-based write request.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

Claims (5)

A method for guaranteeing a recording order of data written to storage by a host, the method comprising:
Invoking at least one order guarantee write request for guaranteeing a write order in the storage, for at least one block data that is data corresponding to an updated data block;
Invoking a barrier-based synchronization command which is a command to cause the at least one block data to be written to the storage prior to subsequent block data called after the at least one ordered write request is invoked;
Invoking a barrier-based write request for dummy data to support guaranteeing a write order in the storage and setting breakpoints in the write order;
Barrier command delivery method using a dummy input and output request comprising a. A method for guaranteeing a recording order of data written to storage by a host, the method comprising:
Invoking at least one order guarantee write request for guaranteeing a write order in the storage, for at least one block data that is data corresponding to an updated data block;
Invoking a barrier-based synchronization command which is a command to cause the at least one block data to be written to the storage prior to subsequent block data called after the at least one ordered write request is invoked;
Determining whether the at least one order-guaranteed write request remains in an IO scheduler queue used to construct a write command from a write request; And
If the at least one write request remains, invoking a barrier-based write request for dummy data to support guaranteeing the write order in the storage and setting breakpoints in the write order;
Barrier command delivery method using a dummy input and output request comprising a. The method of claim 2,
If all or part of the at least one write request remains, changing a write request consisting of the last write command of the at least one write request to a barrier-based write request
Barrier command delivery method using a dummy input and output request characterized in that it further comprises. Storage incorporating software for operating a file system supporting guarantee of recording order; And
A processor for operating software mounted on the storage;
The processor is
Calling at least one order guarantee write request for guaranteeing the write order in the storage, for at least one block data that is data corresponding to an updated data block,
Invoking a barrier-based synchronization command that is a command that causes the at least one block data to be written to the storage prior to subsequent block data called after the at least one ordered write request is invoked,
Determine if the at least one order-guaranteed write request remains in an IO scheduler queue used to construct a write command from a write request,
If the at least one write request does not remain, a dummy input / output request is characterized in that it calls a barrier-based write request for guaranteeing the write order in the storage and setting a break point of the write order for the dummy data. A computing device that carries barrier commands. The method of claim 4, wherein
The processor is
If all or part of the at least one write request remains, the barrier command using the dummy input / output request may be changed into a barrier-based write request. Computing device to deliver.

Images