Valid page threshold based garbage collection for solid state drive
A method for garbage collection in a solid state drive (SSD) includes determining whether the SSD is idle by a garbage collection module of the SSD; based on determining that the SSD is idle, determining a victim block from a plurality of memory blocks of the SSD; determining a number of valid pages in the victim block; comparing the determined number of valid pages in the victim block to a valid page threshold; and based on the number of valid pages in the victim block being less than the valid page threshold, issuing a garbage collection request for the victim block.
BACKGROUND
This disclosure relates generally to computer memory devices, and more particularly to garbage collection in a solid state drive (SSD).
SSDs, which include flash memory devices, are rewritable non-volatile memory devices that use electronic circuitry for storing data. SSDs may be used for many types of computer memory, for example, local computer hard drives, peripheral storage, or storage for enterprise applications. SSDs offer relatively good bandwidth and random I/O (input/output) performance along with a relatively good lifespan due to an absence of moving parts.
SSD memory is organized in memory blocks. Each block includes a plurality of pages (a page is a relatively small unit of memory). Read and write operations in a SSD may be performed at the page level; however, erase operations may only be performed at the memory block level. Erase operations may also take more time to complete than read or write operations. In order to overwrite data (i.e., replace old data with updated data) that is stored in a memory block in a SSD, an erase operation of the entire memory block that holds the data to be overwritten must be performed. In order to overwrite data in a SSD, a writing technique referred to as "write out of place" may be applied, in which the updated data is written to a free (i.e., empty) page in the SSD memory instead of writing the updated data to the same page where old data resides. The page containing the old data is then marked as invalid page. An invalid page remains invalid until the entire memory block in which the invalid page is located is erased.
During operation of the SSD, garbage collection is performed to maintain a free block pool. The free block pool holds memory blocks that contain free pages that are available for writing new data. The free blocks are reclaimed from memory blocks that may be filled with both valid and invalid data. The garbage collection process first identifies a victim block for reclaiming. Any valid pages (i.e., pages containing data that has not been overwritten) residing in the victim block are copied to another memory block, and the victim block is erased, such that the entire victim block is now a free block for the free block pool. The garbage collection process requires read and write operations, in addition to the erase operation, that may compete with regular user read and write operations for access to the same memory and controller resources. Overly frequent garbage collection may increase write amplification in the SSD, which may increase the power consumption of the SSD and eventually decrease the lifespan of the SSD.
SUMMARY
In one aspect, a method for garbage collection in a SSD includes determining whether the SSD is idle by a garbage collection module of the SSD; based on determining that the SSD is idle, determining a victim block from a plurality of memory blocks of the SSD; determining a number of valid pages in the victim block; comparing the determined number of valid pages in the victim block to a valid page threshold; and based on the number of valid pages in the victim block being less than the valid page threshold, issuing a garbage collection request for the victim block.
In another aspect, a SSD includes a plurality of memory blocks, each of the plurality of memory blocks comprising a respective plurality of pages; and a SSD controller, the SSD controller comprising a garbage collection module, the garbage collection module being configured to perform a method including determining whether the SSD is idle; based on determining that the SSD is idle, determining a victim block from the plurality of memory blocks; determining a number of valid pages in the victim block; comparing the determined number of valid pages in the victim block to a valid page threshold; and based on the number of valid pages in the victim block being less than the valid page threshold, issuing a garbage collection request for the victim block.
Additional features are realized through the techniques of the present exemplary embodiment. Other embodiments are described in detail herein and are considered a part of what is claimed. For a better understanding of the features of the exemplary embodiment, refer to the description and to the drawings.
DETAILED DESCRIPTION
Embodiments systems and methods for valid page threshold based garbage collection for a SSD are provided, with exemplary embodiments being discussed below in detail. Garbage collection in a SSD may free up blocks in the SSD without significantly increasing write amplification in the SSD through application of a valid page threshold and an adjustable timeout period to the garbage collection process. When the SSD is determined to be idle, a victim block is determined. Once the victim block is determined, it is determined whether the number of valid pages in the victim block is less than a valid page threshold value. If the number of valid pages in the victim block is less than the valid page threshold, a garbage collection request is issued for the victim block, and the length of the timeout period may also be decreased. However, if the number of valid pages in the victim block is greater than the valid page threshold, the victim block is not garbage collected, and the length of the timeout period may be increased. Application of the valid page threshold and adjustable timeout period to garbage collection may reduce write amplification (which comprises a ratio of the total number of writes in the SSD, both user-generated and SSD-generated, to the number of user writes) in the SSD while ensuring a sufficient free block pool size for good SSD operation. Also, garbage collection operations may be confined to idle time in the SSD, reducing competition for resources between garbage collection and user read/write requests.
The determination that the SSD is idle may be performed by detecting a first idle condition in the SSD, starting the timeout period in response to detecting the first idle condition, and, if a second idle condition is detected in the SSD at the end of the timeout period, determining that the SSD is idle. The first and second idle condition may be determined based on a request queue of the SSD being empty. The duration of the timeout period affects the frequency of garbage collection operations in the SSD. A relatively long timeout period reduces the number of opportunities to grow the free block pool, and may be appropriate in circumstances in which the free block pool is currently relatively large. Because user read and write operations may be processed by the SSD during the timeout period, a relatively long timeout period also increases the chances of identifying a victim block with a number of valid pages that is less than the valid page threshold for garbage collection after the detection of the second idle condition. A relatively short timeout period may grow the free block pool relatively quickly; however, the load on the SSD controller may be increased by more frequent garbage collection operations, with possible detrimental consequences on the SSD controller's response time to other tasks and the controller's energy consumption. Therefore, the length of the timeout period may be adjusted during the garbage collection process based on the state of the SSD, as indicated by the number of valid pages in the determined victim block. A victim block having a number of valid pages that is less than the valid page threshold may trigger a decrease in the timeout period, while a victim block having a number of valid pages that is greater than the valid page threshold may trigger an increase in the timeout period.
The timeout period may be adjusted in any appropriate manner; for example, the timeout period may be increased or decreased exponentially in some embodiments. For example, the timeout duration may be doubled if the determined victim block has a number of valid pages less than the valid page threshold, and the timeout period may be halved if the determined victim block has a number of valid pages greater than the valid page threshold. Any appropriate factor may be used to exponentially adjust the timeout period in various embodiments. Adjustment of the timeout period may be lower-bounded by a timeout minimum value, and upper-bounded by a timeout maximum value. The timeout minimum and maximum values may comprise predetermined values in the SSD. The timeout period may be initially set to the timeout minimum value at startup of the SSD in some embodiments.
In some embodiments, valid page threshold based garbage collection may be disabled when a number of free blocks in the SSD is greater than a threshold T1, and enabled when a number of free blocks in the SSD is less than the threshold T1, so that the free block pool may be increased when the number of free blocks in the SSD is less than a free block threshold T1. In further embodiments, valid page threshold based garbage collection may be initially disabled at startup of the SSD, and only turned on when the number of free blocks in the SSD drops below T1. In yet further embodiments, valid page threshold based garbage collection may be disabled if the number of free blocks in the SSD drops below a second free block threshold T2 wherein T2 is less than T1, such that if the number of free blocks in the SSD drops below the second threshold T2, garbage collection operations may be given a higher priority with respect to user read/write operations. Garbage collection that is performed based on thresholds T1 and/or T2, which may be used in conjunction with some embodiments of valid page threshold based garbage collection is described in European Patent Application No. 11191378.6, which is assigned to International Business Machines Incorporated and which is herein incorporated by reference in its entirety. In other embodiments, valid page threshold based garbage collection may operate regardless of the number of free blocks currently available in the SSD; i.e., in such embodiments, the valid page threshold based garbage collection may be triggered solely based on the determination that the SSD is idle.
FIG. 1 illustrates an embodiment of a SSD 100 including a garbage collection module 103 that implements valid page threshold based garbage collection. The SSD 100 includes a SSD controller 101, which comprises a request queue 102, a garbage collection module 103, and a processor 104. The processor 104 executes read, write, and erase requests from the request queue 102. The requests in the request queue 102 comprise user read/write requests, and garbage collection requests, which include read, write and erase operations. An idle condition may be detected in the SSD 100 by garbage collection module 103 if there are no waiting requests in the request queue 102. The SSD 100 further includes a SSD memory 105, which includes a plurality of memory blocks 106A-N. Each of the plurality of memory blocks 106A-N comprise a plurality of pages. Any appropriate number of memory blocks 106A-N may be included in a SSD memory such as SSD memory 105, and each of the memory blocks 106A-N may include any appropriate number of pages. During operation of the SSD 100, pages in the memory blocks 106A-N may be invalidated by the "write out of place" data overwriting technique. Valid pages in the memory blocks 106A-N comprise pages that hold data that has not been overwritten, and free blocks of the memory blocks 106A-N comprise memory blocks that contain empty pages that may accept writes of new data. The garbage collection module 103 determines expiration of the timeout period, adjusts the timeout period, determines victim blocks from memory blocks 106A-N, and issues garbage collection requests into request queue 102 according to the method 200 illustrated in FIG. 2, which is discussed below. In some embodiments, the SSD 100 may include a plurality of counters (not shown), each counter associated with a respective memory block 106A-N, that each indicate a number of valid pages in the counter's associated memory block; in such an embodiment, these counters may be used to determine the victim block, as is discussed below with respect to block 205.
FIG. 2 illustrates an embodiment of a method 200 for valid page threshold based garbage collection for a SSD, such as SSD 100 of FIG. 1. Method 200 may be implemented in garbage collection module 103. First, in block 201, a first idle condition in SSD 100 is detected. The first idle condition may be detected based on the request queue 102 being empty. When a first idle condition is detected in block 201, flow proceeds to block 202, in which the garbage collection module 103starts the timeout period in response to detection of the first idle condition. Flow then proceeds to block 203, in which it is determined if the timeout period has expired, and then, when it is determined that the timeout period has expired in block203, flow proceeds to block 204, and it is determined whether a second idle condition is detected in the SSD 100. User read/write requests may be issued into the request queue 102 and executed by the processor 104 during the elapsing of the timeout period during blocks 202 and 203; the second idle condition is detected in block 204 based on the request queue 102 being empty at the time of expiring of the timeout period. If, in block 204, the second idle condition is not detected, flow of method 200 returns to block 201, and the garbage collection module 103 awaits another idle condition.
If, in block 204, the second idle condition is detected, flow of method 200 proceeds to block 205, in which the garbage collection module 103 determines a victim block from memory blocks 106A-N. The victim block may be determined in any appropriate manner. In some embodiments, the memory blocks may be directly examined by the garbage collection module103 to determine the victim block. In some embodiments, only a subset of the memory blocks 106A-N in the SSD memory105 may be examined during the determination of the victim block by garbage collection module 103. In other embodiments, the SSD 100 may include a plurality of counters, each counter associated with a respective memory block 106A-N, that each indicate a number of valid pages in the counter's associated memory block; in such an embodiment, these counters may be used to determine the victim block. For example, a memory block of memory blocks 106A-N whose counter has reached the valid page threshold may be selected as the victim block by the garbage collection module in block 205.
After the victim block is determined in block 205, flow proceeds to block 206 of method 200, in which it is determined if the number of valid pages in the victim block is less than the valid page threshold. If the number of valid pages in the victim block is determined to be less than the valid page threshold in block 206, a garbage collection request (comprising copying the valid pages in the victim block to another block, and erasing the entire victim block) for the victim block is issued by garbage collection module 103 into request queue 102 in block 207, and flow then proceeds to block 208, in which the timeout period is decreased, subject to the timeout period minimum. Flow then proceeds from block 208 back to block 201. If the number of valid pages in the victim block is determined to be greater than the valid page threshold in block 206, the victim block is not garbage collected, and flow proceeds to block 209, in which the timeout period is increased, subject to the timeout period maximum. Flow then proceeds from block 209 back to block 201.
The value of the valid page threshold, to which the number of valid pages in the victim block is compared in block 206 ofFIG. 2, may be selected in any appropriate manner. In some embodiments, the valid page threshold may be a predetermined value. In some embodiments, the valid page threshold may be equal to (1−1/A)*c, where A is a target write amplification value for the SSD, and c is the number of pages per memory block. A safety margin may be factored into the value of the valid page threshold by reducing the value given by (1−1/A)*c by an appropriate percentage in some embodiments. In other embodiments, the garbage collection module 103 may compute an average number of valid pages per garbage collection operation that is performed by the garbage collection module. In such embodiments, the valid page threshold may be set equal to a most recently computed average of the number of valid pages per garbage collection operation, or, if a safety margin is desired, an appropriate percentage thereof. In further embodiments, the valid page threshold may be dynamically adapted to the current number of free blocks in the SSD. In such embodiments, when there is a relatively large numbers of free blocks, garbage collecting to free up additional blocks is less critical and may be performed only if the price paid, i.e., the number of relocated valid pages, is relatively small. Therefore, the valid page threshold may decrease (or at least not increase) based on the number of free blocks in the free block pool
The timeout period minimum and maximum, which give the lower and upper bounds for increasing and decreasing of the timeout period in blocks 208 and 209 of method 200 of FIG. 2, may be selected in any appropriate manner. The amount by which the timeout period is increased and decreased in blocks 208 and 209 may also be selected in any appropriate manner. The timeout period may initially be set equal to the timeout period minimum at startup of the SSD 100 in some embodiments. In some embodiments, the timeout period minimum may be about a millisecond, and the timeout period maximum may be about a second. The timeout period may be increased and decreased exponentially by any appropriate factor in blocks 208 and 209 of method 200; for example, the timeout period may be doubled in block 209, and halved in block 208
FIG. 3 illustrates an example of a computer 300 which may be utilized by exemplary embodiments of a SSD implementing valid page threshold based garbage collection. Various operations discussed above may utilize the capabilities of the computer 300. One or more of the capabilities of the computer 300 may be incorporated in any element, module, application, and/or component discussed herein. For example, embodiments of a SSD implementing valid page threshold based garbage collection may be incorporated into memory 320, and/or a SSD implementing valid page threshold based garbage collection may comprise a peripheral device that communicates with computer 300 via input/output (I/O) devices370.
SRC=http://www.google.com/patents/US20140032817
Valid page threshold based garbage collection for solid state drive的更多相关文章
- Java Garbage Collection Basics--转载
原文地址:http://www.oracle.com/webfolder/technetwork/tutorials/obe/java/gc01/index.html Overview Purpose ...
- 2.5 – Garbage Collection 自动垃圾回收 Stop-the-world vs. incremental vs. concurrent 垃圾回收策略
2.5 – Garbage Collection 自动垃圾回收 Lua 5.3 Reference Manual http://www.lua.org/manual/5.3/manual.html# ...
- Garbage Collection Optimization for High-Throughput and Low-Latency Java Applications--转载
原文地址:https://engineering.linkedin.com/garbage-collection/garbage-collection-optimization-high-throug ...
- On-the-fly Garbage Collection: an Exercise in Cooperation
On-the-fly Garbage Collection: an Exercise in Cooperation - Microsoft Research https://www.microsoft ...
- [转] 分代垃圾回收的 新旧代引用问题(原标题:Back To Basics: Generational Garbage Collection)
原文链接: https://blogs.msdn.microsoft.com/abhinaba/2009/03/02/back-to-basics-generational-garbage-colle ...
- Fundamentals of Garbage Collection
[Fundamentals of Garbage Collection] 1.Reclaims objects that are no longer being used, clears their ...
- Advanced .NET Debugging: Managed Heap and Garbage Collection(转载,托管堆查内存碎片问题解决思路)
原文地址:http://www.informit.com/articles/article.aspx?p=1409801&seqNum=4 Debugging Managed Heap Fra ...
- 深入理解Java中的Garbage Collection
前提 最近由于系统业务量比较大,从生产的GC日志(结合Pinpoint)来看,需要对部分系统进行GC调优.但是鉴于以往不是专门做这一块,但是一直都有零散的积累,这里做一个相对全面的总结.本文只针对Ho ...
- Why you need to understand garbage collection
Why you need to understand garbage collection I’ve been interviewing lots of C# developers recently, ...
随机推荐
- Spring Boot系列二 Spring @Async异步线程池用法总结
1. TaskExecutor Spring异步线程池的接口类,其实质是java.util.concurrent.Executor Spring 已经实现的异常线程池: 1. SimpleAsyncT ...
- IIS7配置PHP图解
IIS6整合PHP详解:http://zhidao.zgsj.com/article/8/2011118142648.shtml PHP5.2.17 官方下载: http://windows.php. ...
- Identity of indiscernibles(不可分与同一性)
不可区分的对象或者实体,全部性质都相同.也就是说,实体 x 与实体 y是同一的,如果对 x 成立的全部断言,对 y 同样成立(反之亦然). 注意区分如下的两个概念: 同一性的不可分性(The indi ...
- P2P网贷第三方托管模式存在5大缺陷,托管机构才是最大赢家
1.注册开户需要2次,用户体验很差劲儿. 理财人和借款人,首先在平台注册,然后还要在第三方托管账户注册. 很多类似的地方,用户体验非常差劲. 比如,密码4个. 平台:登录密码.交易密码 ...
- 服务器负载均衡lvs(Linux Virtual Server)
服务器负载均衡lvs(Linux Virtual Server) 一.总结 LVS是Linux Virtual Server的简写,意即Linux虚拟服务器,是一个虚拟的服务器集群系统. 三.Linu ...
- [Angular] Using NO_ERRORS_SCHEMA
For example, when testing container component we might import lots of children components, but we di ...
- HTML5 canvas 指针时钟
<!doctype html> <html> <head></head> <body> <canvas id="> 您 ...
- 【codeforces 742B】Arpa’s obvious problem and Mehrdad’s terrible solution
time limit per test1 second memory limit per test256 megabytes inputstandard input outputstandard ou ...
- php面试题11(边看边复习刚刚讲的)(array_multisort($arr1,$arr2); 用$arr1来排序$arr2。)
php面试题11(边看边复习刚刚讲的)(array_multisort($arr1,$arr2); 用$arr1来排序$arr2.) 一.总结 1.边看边复习刚刚讲的 2.array_multisor ...
- Dynamips GNS3
https://baike.baidu.com/item/dynamips Dynamips的原始名称为Cisco 7200 Simulator,源于Christophe Fillot在2005年8月 ...