KR20110124395A - Method for managing the read and write cache in the system comprising hybrid memory and ssd - Google Patents

Abstract

PURPOSE: A hybrid memory and R/W(Read and Write) cache management method of a system in which is constituted by a hybrid memory and SSD is provided to improve the function of reading and writing by effectively managing a R/W cache considering of the property of the memory and the SSD. CONSTITUTION: A hybrid memory includes a non-volatile memory(130) and a volatile memory(120). A SSD(Solid State Disk)(110) is connected to the hybrid memory through a bus. The SSD is operated ad a storage unit. The hybrid memory is operated as the cache of the SSD. The volatile memory includes a first R/W cache. The non-volatile memory includes a second R/W cache. The first cache data of the volatile memory is selectively eliminated in order to move to the second R/W cache of the non-volatile memory.

Description

METHODS FOR MANAGING THE READ AND WRITE CACHE IN THE SYSTEM COMPRISING HYBRID MEMORY AND SSD}

The present invention relates to a method of managing a read cache and a write cache in a system consisting of a hybrid memory and an SSD, and in particular, a read cache and a write cache for improving system performance by utilizing characteristics of a volatile memory, a nonvolatile memory, and an SSD. It is about how to efficiently manage the same time.

The read and write caches of today's computer systems are composed of volatile memory, allowing read and write performance to be reduced by reducing the number of reads and writes to disk by reading and writing frequently used and frequently read data from the cache instead of the disk. It is becoming. This takes advantage of the volatile memory's read and write performance over disk. However, since volatile memory has a low density and a high price / capacity, configuring a cache using only volatile memory limits the size of available volatile memory, so the amount of data that can be stored in volatile memory is limited. This increases the number of reads and writes to the disk.

In order to solve the above problem, the conventional patent US5343437, US6263398 and the like proposed a two-stage main memory structure using heterogeneous memory that uses volatile memory as a primary memory and nonvolatile memory as a secondary main memory. Nonvolatile memory has the disadvantages of poor write performance and limited number of writes compared to volatile memory, but read performance is similar to volatile memory, and its density is higher than volatile memory. Therefore, the main memory structure consisting of volatile and nonvolatile memory can have a larger capacity than the main memory structure consisting only of volatile memory, which can cache a lot of data, thereby reducing the number of writes and reads to the disk. There is an advantage.

Accordingly, a technical problem to be achieved in the present invention relates to a method of managing a read cache and a write cache in a system consisting of a hybrid memory and an SSD, the capacity of the main memory through a hybrid structure with non-volatile memory as well as volatile memory In this structure, the read cache and write cache management method and the storage device using SSD, the next generation storage device rather than the existing hard disk, utilize the characteristics of the SSD and read cache and write cache management method through the method of the whole system. To improve performance.

The read cache management method improves the read cache performance by selectively removing unnecessary data when moving data from volatile memory to nonvolatile memory, thereby reducing the number of unnecessary data writes to the nonvolatile memory, thereby efficiently saving the space of the nonvolatile memory. It can be used as a product and its life can be extended.

In the write cache management method, considering the multi-channel structure of the SSD, it manages as superpages (pages * channels) in the write cache, thereby improving performance by increasing parallelism (Parallelis) in write operations to the SSD. When a superpage is moved from nonvolatile memory to volatile memory, copying only frequently used pages, instead of moving the entire page, reduces unnecessary page movement time and improves the performance of the write cache. This reduces the number of extra writes and increases the lifespan.

In addition, the size of the read cache and write cache changes depending on the workload characteristics, so that the cache can be utilized to the maximum.

A memory system according to an embodiment of the present invention, a hybrid memory including a volatile memory and a nonvolatile memory; And an SSD connected via a hybrid memory and a bus, wherein the SSD operates as a storage device and the hybrid memory operates as a cache of the SSD.

Preferably, the memory system includes volatile memory including a first read cache and a first write cache, and nonvolatile memory includes a second read cache and a second write cache.

Preferably, in the memory system, data in the first read cache of volatile memory is moved to the second read cache of nonvolatile memory by selective removal.

Preferably, in the memory system, the data moved by selective removal is data pre-fetched from the SSD into the first read cache.

Advantageously, in a memory system, frequently used data is moved from said first read cache or said second read cache to one of said first write cache and said second write cache of said non-volatile memory by selective removal.

Preferably, in the memory system, the data moved by selective removal is sequential data.

Preferably, in the memory system, in order to increase the write performance from the second write cache of the nonvolatile memory to the SSD, the data in the second write carry is managed in units of superpages in consideration of the internal structure of the SSD.

Preferably, in the memory system, only the most frequently used pages of the super pages of the second write cache are copied to the volatile memory.

Preferably, in the memory system, the size of the super page is defined as the product of the number of channels within the SSD and the page, which is a basic operation unit of reading and writing of the SSD.

Advantageously, in a memory system, increase the size of the first read cache and the second read cache if there are many read operations based on the workload, and increase the size of the first write cache and the second write cache if there are many write operations. .

According to an embodiment of the present invention, a method of managing a memory system including a volatile memory, a nonvolatile memory, and an SSD may include: receiving a disk read command from an external device; Writing data stored in the SSD to a read cache of the volatile memory in response to the disk read command; Checking whether there is free space in the read cache of the volatile memory; And if there is no free space in the read cache, moving the prefetched data to the read cache of the nonvolatile memory.

According to an embodiment of the present invention, a method of managing a memory system including a volatile memory, a nonvolatile memory, and an SSD may include: receiving a disk read command from an external device; Writing data stored in the SSD to a read cache of the volatile memory in response to the disk read command; Checking whether there is free space in the read cache of the volatile memory; And when there is no free space in the read cache, moving the frequently used data to the read cache of the nonvolatile memory.

According to an embodiment of the present invention, a method of managing a memory system including a volatile memory, a nonvolatile memory, and an SSD may include: receiving a disk read command from an external device; Writing data stored in the SSD to a read cache of the volatile memory in response to the disk read command; Checking whether there is free space in the read cache of the volatile memory; And when there is no free space in the read cache, moving the frequently used data to the read cache of the nonvolatile memory.

According to an embodiment of the present invention, a method of managing a memory system including a volatile memory, a nonvolatile memory, and an SSD may include: receiving a disk read command from an external device; Writing data stored in the SSD to a read cache of the volatile memory in response to the disk read command; Checking whether there is free space in the read cache of the volatile memory; And when there is no free space in the read cache, moving the sequential data to the read cache of the nonvolatile memory.

According to an embodiment of the present invention, a method of managing a memory system including a volatile memory, a nonvolatile memory, and an SSD may include: receiving a disk write command from an external device; Detecting a frequency of use of pages included in a write cache of the nonvolatile memory; And copying the frequently used pages from the write cache of the nonvolatile memory to the write cache of the volatile memory based on the usage frequency.

As described above, the present invention proposes an efficient management of the read cache and the write cache in consideration of the characteristics of each memory and SSD in a system consisting of a hybrid memory and an SSD, consequently contributed to improving the read and write performance of the system. .

In order to take advantage of the fast read and write capability, which is characteristic of volatile memory, the read cache is used to store cache data and prefetch data, and the write cache optimizes cache performance by storing only frequently used data.

Nonvolatile memory has the characteristics that it has a larger capacity at the same price than volatile memory and that read performance is similar to volatile memory, but to overcome the disadvantages of poor write performance and limited number of writes, Selective elimination techniques in the cache, super page management in the write cache, and frequently used data movement can address these shortcomings.

SSD read and write units are pages (2KB), and internally composed of multiple channels to increase parallelism, it manages them as super pages in the write cache to take advantage of this property when writing from hybrid memory to disk. It can improve performance.

It also takes advantage of the fact that read caches and write caches are managed in one memory, which makes it possible to manage the cache in memory more efficiently by using size conversion based on workload characteristics.

1 illustrates an example of an overall system structure including a hybrid memory and an SSD according to the present invention.
2 is a diagram illustrating an example of a method of managing a read cache in a hybrid memory according to an embodiment of the present invention.
3 is a diagram illustrating an example of a method of managing a write cache in a hybrid memory according to the present invention.
4 is a diagram illustrating an example of adaptive size conversion of the read cache and the write cache in the volatile memory 120 and the nonvolatile memory 130 according to the present invention.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. However, the following examples are merely to illustrate the present invention is not limited to the contents of the present invention.

1 illustrates an example of an overall system structure including a hybrid memory and an SSD according to the present invention, and includes a central processing unit 100, a hybrid memory, and a solid state disk (SSD) 110. The hybrid memory includes a volatile memory ( 120 and the nonvolatile memory 130. The volatile memory 120 and the nonvolatile memory 130 may be connected through a host bus, and the host bus may be connected to the SSD 110 through the bridge 140.

The volatile memory 120 has the advantage of fast writing and reading speed, but has a disadvantage of being expensive for capacity. The nonvolatile memory 130 has a slower writing speed and a limited number of writes than the volatile memory 120. It has the disadvantage of high density, low price for capacity, and similar read speed to volatile memory.

SSD 110 is a NAND flash memory-based storage device to replace the hard disk has the advantage that the performance is better than the hard disk for any request, and the internal structure is disclosed that the performance can be predicted for read and write operations Has

Since the SSD 110 is composed of several channels, each channel can be read and write independently, and thus the performance of the SSD can be increased by using these channels simultaneously. If you manage the read cache and write cache by using each characteristic in the structure of hybrid memory and SSD, the performance of the entire system can be improved.

2 is a diagram illustrating an example of a method of managing a read cache in a hybrid memory according to an embodiment of the present invention.

The purpose of using the read cache is to improve read performance by storing data that is frequently read in memory because reading data from disk is slower than memory. In the present invention, since the SSD is used instead of the disk, the read cache serves as a storage space for temporarily storing data when the SSD is accessed from the central processing unit. The read cache may be in volatile memory 120 and / or nonvolatile memory 130 of hybrid memory.

The read cache management in the hybrid memory is managed by being divided into the volatile memory 120 and the nonvolatile memory 130. First, data that is frequently read is first stored in the volatile memory 120 according to the management method of the read cache. In addition, in the read cache, prefetch data that reads data from the disk in anticipation of the data to be read is present. The data is stored in the volatile memory 120 and the data is moved to the nonvolatile memory 130 when the volatile memory 120 is full. At this time, all data from the volatile memory 120 to the nonvolatile memory 130 are transferred. Does not move but removes data that does not need to be stored in the volatile memory 120 and moves to the nonvolatile memory 130 through a method called selective removal.

The reason for this selective removal is that the nonvolatile memory 130 has a relatively slow write speed and a limited number of writes compared to the volatile memory 120, and therefore, storing unnecessary data in the nonvolatile memory 130 has a negative effect on performance. Because it affects.

The selective removal method can have three embodiments.

The first method is to selectively remove prefetched data. Prefetch means that when the hybrid memory controller (not shown) receives a data read command from the central processing unit, data that is expected to be read later is read from the SSD 110 into the volatile memory 120 in advance. .

Since the data prefetched from the volatile memory 120 is not read frequently, it is pushed to the nonvolatile memory 130, and therefore stored in the nonvolatile memory 130 has a low probability of being read. Therefore, since this causes unnecessary write operations, selective removal of this prefetched data is better in terms of stability of the nonvolatile memory 130.

The second method is to selectively remove write-intensive data. In other words, this method selectively removes data with high access frequency. Frequently used data is not a problem because there is no limit on the number of writes in volatile memory (120). However, since nonvolatile memory (130) has a limit on the number of writes, a method of selectively removing frequently used data improves the performance of the read cache. Would be better for you.

However, in case of prefetch data, the data is moved to the nonvolatile memory. However, the data that is used frequently is moved to the write cache of the volatile memory so that when a write operation occurs, the data is written to the write cache instead of the disk.

Finally, the method of selectively removing sequential data.

Because sequential data is less likely to be read than random data, caching them unnecessarily takes up storage space in the read cache. Accordingly, when data is moved from the volatile memory 120 to the nonvolatile memory 130, the sequential data may be selectively removed to relatively increase the storage space of the nonvolatile memory 130 and prevent unnecessary write operations.

3 is a diagram illustrating an example of a method of managing a write cache in a hybrid memory according to the present invention.

Like the read cache, the write cache saves frequently used data in memory and increases performance by reducing the number of writes to disk.

The write cache is also managed by dividing it into volatile memory (and nonvolatile memory). First, in nonvolatile memory, it is managed as a superpage (number of pages * channels). This is to improve performance when used in.

Since the SSD 110 is composed of several channels, parallel operation is possible, and the parallelism can be maximized by managing data with a superpage. Therefore, the size of the superpage may vary according to the number of channels of the SSD 110.

When a non-volatile memory 130 is managed as a super page and frequently used pages are copied to the volatile memory 120.

This is to reduce the number of write operations when frequently used pages are in the nonvolatile memory 130. In addition, by copying a page other than the super page, the moving overhead is reduced and the write cache management in the volatile memory 120 can be more efficiently performed.

When the copy process is performed as described above, the copied page is simultaneously present in both the volatile memory 120 and the nonvolatile memory 130.

When the contents of the write cache of the volatile memory 120 are updated, when the write cache space of the volatile memory 120 is insufficient and needs to be moved to the nonvolatile memory 130 or the disk later, the nonvolatile memory 130 If a superpage with this page exists in the nonvolatile memory 130, the update is performed. If the nonpage exists in the nonvolatile memory 130, the superpage is read from the disk and the superpage is read from the nonvolatile memory 130. Only the page will be updated.

The reason for copying rather than moving pages from nonvolatile memory 130 to volatile memory 120 should be written back to nonvolatile memory 130 later when there is no storage space in volatile memory 120. This is to reduce extra write operations.

As such, when duplicate data exists in the volatile memory 120 and the nonvolatile memory 130, there is a disadvantage in utilizing the write cache storage space, but the additional capacity is larger than this overhead because the capacity of the nonvolatile memory 130 is large. It is thought that the write operation overhead is greater.

Therefore, by performing the write cache management, the write performance to the SSD 110 is also increased, unnecessary write operations to the nonvolatile memory 130 can be reduced, and the volatile memory 120 is used as a write cache of a frequently used page. There is an advantage that can increase the performance.

4 is a diagram illustrating an example of adaptive size conversion of the read cache and the write cache in the volatile memory 120 and the nonvolatile memory 130 according to the present invention.

Since the read cache and the write cache coexist in the volatile memory 120 and the nonvolatile memory 130, the size conversion thereof will greatly affect the performance. Therefore, first of all, consider the change in size according to the workload characteristics. For a workload with many write operations, it is possible to allocate a large write cache and a small read cache.

Conversely, for workloads with many read operations, cache size allocation may be required. In addition, performance may be changed by changing the cache size in the volatile memory 120 and the cache size in the nonvolatile memory 130.

As described above, those skilled in the art to which the present invention pertains will understand that the present invention may be implemented in other specific forms without changing the technical spirit or essential features.

Therefore, the embodiments described above are to be understood in all respects as illustrative and not restrictive, and the scope of the present invention is indicated by the following claims rather than the above description, and the meaning and scope of the claims and All changes or modifications derived from the equivalent concept should be interpreted as being included in the scope of the present invention.

100: central processing unit
120: volatile memory
130: nonvolatile memory
140: the bridge
110: SSD (Solid State Disk)

Claims (15)

A hybrid memory including a volatile memory and a nonvolatile memory; And
An SSD connected via a bus and the hybrid memory,
Wherein said SSD operates as a storage device and said hybrid memory operates as a cache of said SSD. The method of claim 1,
The volatile memory includes a first read cache, and a first write cache,
And the non-volatile memory includes a second read cache and a second write cache. The method of claim 1,
Data in the first read cache of the volatile memory is moved to a second read cache of the nonvolatile memory by selective removal. The method of claim 3, wherein
And the data moved by the selective removal is data pre-fetched from the SSD into the first read cache. The method of claim 1,
Frequently used data is moved from the first read cache or the second read cache to one of the first write cache and the second write cache of the nonvolatile memory by selective removal. The method of claim 3, wherein
And the data moved by the selective removal is sequential data. The method of claim 2,
In order to increase the write performance of the non-volatile memory from the second write cache to the SSD, the data in the second write carry is managed in units of superpages in consideration of the internal structure of the SSD. The method of claim 7, wherein
And copy only the most frequently used pages of the super write pages of the second write cache to volatile memory. The method according to claim 7 or 8,
The size of a super page is defined as the product of the number of pages within the SSD and the page, which is the basic unit of operation for reading and writing the SSD. The method of claim 2,
Based on the workload, increase the size of the first read cache and the second read cache if there are many read operations, and increase the size of the first write cache and the second write cache if there are many write operations. A method of managing a memory system including volatile memory, nonvolatile memory, and SSD,
Receiving a disk read command from the outside;
Writing data stored in the SSD to a read cache of the volatile memory in response to the disk read command;
Checking whether there is free space in a read cache of the volatile memory;
If there is no free space in the read cache, moving prefetched data to a read cache of the nonvolatile memory. A method of managing a memory system including volatile memory, nonvolatile memory, and SSD,
Receiving a disk read command from the outside;
Writing data stored in the SSD to a read cache of the volatile memory in response to the disk read command;
Checking whether there is free space in a read cache of the volatile memory; And
If there is no free space in the read cache, moving the frequently used data to the read cache of the nonvolatile memory. A method of managing a memory system including volatile memory, nonvolatile memory, and SSD,
Receiving a disk read command from the outside;
Writing data stored in the SSD to a read cache of the volatile memory in response to the disk read command;
Checking whether there is free space in a read cache of the volatile memory; And
If there is no free space in the read cache, moving the frequently used data to the read cache of the nonvolatile memory. A method of managing a memory system including volatile memory, nonvolatile memory, and SSD,
Receiving a disk read command from the outside;
Writing data stored in the SSD to a read cache of the volatile memory in response to the disk read command;
Checking whether there is free space in a read cache of the volatile memory; And
If there is no free space in the read cache, moving the sequential data to the read cache of the nonvolatile memory. A method of managing a memory system including volatile memory, nonvolatile memory, and SSD,
Receiving a disk write command from the outside;
Detecting a frequency of use of pages included in the write cache of the nonvolatile memory; And
Copying a high usage page based on the usage frequency from a write cache of the nonvolatile memory to a write cache of the volatile memory.

Images