Solid-state storage management for a system, the management including establishing, externally to a solid-state storage board, a correspondence between a first logical address and a first physical address on solid-state storage devices located on the solid-state storage board. The solid-state storage devices include a plurality of physical memory locations identified by physical addresses. The correspondence between the first logical address and the first physical address is accepted by the solid-state storage board. The correspondence between the first logical address and the first physical address is stored in a location on a solid-state memory device that is accessible by an address translator module, the address translator module and the solid-state memory device located on the solid-state storage board. The first logical address is translated to the first physical address by the address translator module based on the previously established correspondence between the first logical address and the first physical address.
BACKGROUND
The present invention relates to computer storage, and more specifically, to solid-state storage management.
Solid-state drives (SSDs) are data storage devices that use solid-state memory to store persistent data. SSDs are most commonly connected to a host computer via a high speed link. SSDs usually implement some form of management of the storage medium, i.e., the underlying solid-state storage devices. Such management usually includes translating from a logical address space into a physical address space. Many SSDs use "not-and" (NAND) based flash memory which must be erased in blocks before being rewritten with new data. Address translation is commonly used with NAND flash devices due to the difficulty of erase operations.
In some contemporary systems that use NAND flash devices, all of the management functionality is performed on the host computer. In these systems, the address translation table is placed in the main memory of the host computer such that during a read operation the host computer determines the physical address corresponding to the logical read address (address translation). During a write operation, the host computer decides the physical write address to which the data is written (physical data placement), and logs the logical-to-physical address correspondence in the address translation table. A shortcoming of this approach is that it requires the expenditure of computational cycles and main memory at the host in order to perform the address translation. Thus, it is well-suited for computer systems with large amounts of dynamic random access memory (DRAM) and processing capabilities, and is not well suited for lower cost systems with limited amounts of DRAM and processing capabilities, or for large-scale data-centric systems with high ratios between storage and system memory.
Other contemporary systems place all of the management functionality on the physical SSD board. In this case, SSD does both data placement and address translation. A shortcoming of this approach is that it suffers in terms of cost/performance because it does not leverage the host processor and instead requires another processor to be incorporated into the SSD.
SUMMARY
An embodiment is a system that includes a main board and a solid-state storage board separate from the main board. The main board is coupled to the solid-state storage board via an interface bus. The solid-state storage board includes a solid-state memory device and solid-state storage devices. The solid-state storage devices include a plurality of physical memory locations identified by physical addresses. The system is configured to perform a method that includes establishing a correspondence between a first logical address and a first physical address on the solid-state storage devices. The establishing is by a software module located on the main board. The correspondence between the first logical address and the first physical address is stored in a location on the solid-state memory device located on the solid-state storage board. The method also includes translating the first logical address into the first physical address. The translating is performed by an address translator module located on the solid-state storage board and is based on the previously established correspondence between the first logical address and the first physical address.
Other embodiments include a method and computer program product for establishing a correspondence between a first logical address and a first physical address on solid-state storage devices located on solid-state storage board. The solid-state storage devices include a plurality of physical memory locations identified by physical addresses, and the establishing is by a software module located on a main board that is separate from the solid-state storage board. The main board is coupled to the solid-state storage board via an interface bus. The correspondence between the first logical address and the first physical address is stored in a location on a solid-state memory device that is accessible by an address translator module located on the solid-state storage board. The solid-state memory device is located on the solid-state storage board. The first logical address is translated to the first physical address by the address translator module based on the previously established correspondence between the first logical address and the first physical address.
DETAILED DESCRIPTION
Embodiments of the present invention provide high performance solid-state drive (SSD) management solutions for use with low-cost or large-scale data-centric computer systems. Specifically, embodiments include an innovative split of flash memory management complexity between a host system and a SSD board, also referred to herein as a hardware module or solid-state storage board. In particular, a host system processor executes a software module that utilizes system resources to perform a specific subset of the tasks involved in SSD management, whereas the remaining tasks are performed and accelerated directly by the SSD board. The hardware module incorporates a controller sub-module, as well as "not-and" (NAND) flash memory and some high speed memory such as dynamic random access memory (DRAM), phase-change memory (PCM) and/or spin-torque transfer random access memory (STT-RAM). The software module handles complex flash memory management operations including, but not limited to: garbage collection, wear leveling, data placement and scheduling, second-tier error correction coding (ECC), redundant array of independent disks (RAID) management, and flash recovery. On the other hand, the hardware controller module handles simple flash translation layer (FTL) bookkeeping (the FTL is the address translation table in NAND-flash systems) and flash command routing, as well as hardware friendly operations such as RAID acceleration and primary-tier ECC using algebraic codes.
Embodiments described herein have very small footprints on the host system and processor in terms of processing power and host memory. Even the most complex flash bookkeeping operations require only a fraction of the computing power of a typical modern host processor, whereas full flash management has a significant impact on system memory occupancy and, as a consequence system performance. Thus, embodiments can be used in systems with inexpensive memory subsystems and relatively inexpensive processors, or systems where power efficiency is a top priority and most of the memory resources will be claimed by the main user application, or large-scale data-centric systems which use several low power compute nodes and/or micro-servers. In addition, the simple hardware module allows low latency and large bandwidth FTL transactions and delivery of flash commands to the flash memory devices. Thus, even on systems running independent concurrent high throughput memory intensive applications, the system and method allows very high performance to be attained. Further, due to the simplicity of hardware this performance is achieved at relatively low costs. The cost/performance of exemplary embodiments in terms of cost/gigabytes (GB), for example, or cost/bandwidth, is significantly better than conventional solutions. Embodiments described herein are well suited for providing high performance at a low cost for systems with limited available resources.
Turning now to FIG. 1, an embodiment of a flash controller system is generally shown. The system includes a host computer 100 connected or coupled to a hardware module 102 by an interface 104. In an embodiment, the host computer100 is a computer system with a main board and a connection to one or more high-speed buses that serve as the interface104 to the hardware module 102. The host computer 100 includes hardware located on the main board (e.g., a motherboard), such as a processor and memory, that execute software, including software module 106. In an embodiment, the software module 106 is responsible for most of the complex flash controller management functions such as garbage collection, wear leveling, optimizing data placement and scheduling, RAID management, flash disaster recovery and higher layer ECC functionality.
In an embodiment, the hardware module 102 is a physical device and responsible primarily for housing a flash module 110, a controller sub-module 108 and a fast memory module 112. Herein, the terms "hardware module" and "solid-state storage board" are used interchangeably. In an embodiment, the controller sub-module 108 accesses the FTL, which resides in one or more solid-state memory devices such as a DRAM 112 or some other high speed volatile or nonvolatile memory on the hardware module 102. As known in the art, the FTL includes a mapping between logical addresses and physical addresses. The flash module 110 includes flash memory devices (e.g., chips) 118 that reside on the hardware module 102. A flash memory device is an example of one type of solid-state storage device that may be implemented by exemplary embodiments. Other examples of solid-state storage devices that may be implemented include PCM, magnetic RAM and resistive RAM.
The main functions of the hardware module 102 include the efficient translation of logical-to-physical addresses to facilitate reading from the flash memory devices 118, and low latency and high bandwidth delivery of flash commands to the flash memory devices 118. To this end, the hardware module 102 also incorporates interface management functions, which manage a fast memory interface 114 between a fast FTL memory module 112 and the flash module 110. In addition, the hardware module 102 is responsible for some simple operations suitable for hardware acceleration, that run on the controller sub-module 108 such as primary error correction code (ECC) and some RAID acceleration operations. The controller sub-module 108 is connected to the flash module 110 via a flash interface 116. In addition, the hardware module102 facilitates the recovery of physical-to-logical address mapping to enable effective operation of management algorithms, including data placement and garbage collection algorithms, as performed by the software module 106. In an embodiment, the hardware module 102 and software module 106 on the host computer 100 are interconnected using interface 104. In an embodiment, interface 104 is a high-speed host hardware interface, such as a high-speed Peripheral Component Interconnect Express (PCIe) interface.
The embodiment depicted in FIG. 1 for SSD management utilizes an innovative split of flash controller management functionality between the software module 106 (e.g., a driver executing on the host computer 100) and the hardware module 102 (e.g., executing on a card with suitable hardware, such as a field programmable gate array or "FPGA", or a controller application specific integrated circuit or "ASIC"). In an embodiment, hardware for acceleration of specific functionalities such as ECC and a table used for translation between logical addresses and physical addresses are in the hardware module 102, and flash management processes such as data placement, wear leveling and control (e.g., based on a wear measurement), and garbage collection, and metadata such as wear data and page status are in the host computer100. The simplicity of the arrangement and hardware reduces both the cost of the required host computer system and the complexity of the design of hardware module 102. For example, in the embodiment depicted in FIG. 1 the hardware module does not need to implement generic processor core functionality, as commonly done in the art.
The split of the flash controller functionality between the host computer 100 and hardware module 102 allows very high performance to be attained. This is due to the software module 106 on the host computer 100 controlling the complex flash operations, where the host computer 100 has sufficient computing system resources for the operations. This is in contrast to performing complex flash operations entirely on the hardware module 102, where resources on the hardware module 102may cause a bottleneck for complex operations. The arrangement depicted in FIG. 1 also saves costly system resources, such as main memory on the host computer 100, by placing the FTL memory module 112 on the hardware module 102. Thus, embodiments are used to facilitate high performance flash I/O in new applications, such as micro-servers in data-center networking, where system resources such as main memory are constrained, and allow accelerated development due to the simplicity of the hardware module 102. A description of how exemplary flash operations are performed according to embodiments is shown in FIGS. 2-6.
FIG. 2 is a block diagram of a process for performing a read flash operation (also called read logical address flash operation) according to an embodiment. In a first block 200, a user read request for a SSD, such as a flash memory device118, is received by the host computer 100. At block 202, the read request is split into physical read units, where each unit has its own logical address. The requests are then queued to an outbound request queue in block 204. At block 206, the hardware module 102 is checked, either directly, or by inferring its status given the previous history of commands and requests and the capabilities of the hardware module 102 itself, to see if it is cleared to send (CTS) the read requests. The hardware module 102 indicates CTS if it is free to perform or accept commands. If the hardware module 102 is CTS, the requests are sent to the hardware module 102, as shown in block 208. In block 210, if the module is not cleared, the system will wait until the hardware module 102 indicates it is clear. In an embodiment, blocks 200-210 are performed by the host computer 100, and blocks 212-220 are performed by the hardware module 102.
In block 212, the read requests for each of the logical addresses are received. In block 214, the hardware module 102determines if there is an outstanding read request, for example at the same hardware resource (possible hardware resources include a flash module, a memory device or chip, a die, of plane) where the read operation has to be performed. If there are outstanding read requests, the module waits, as shown in block 215. If there are no outstanding read requests, block 216 is performed to read the fast FTL memory module 112 (also referred to as "FTL") at the specified logical address to determine the physical address that corresponds to the logical address. At block 218, a read command is sent to the physical address obtained from the FTL. The FTL performs the translation of the logical address to the physical address on the hardware module 102. In block 220, after the read is completed, the data is delivered along with the logical address by the hardware module 102 to the host computer 100 via the interface 104. In another embodiment, the physical address is also sent to the host computer 100 via the interface.
FIG. 3 is a block diagram of a process for a write flash operation according to an embodiment. At block 300, the host computer 100 receives a user write request. In block 302, the write request is split into physical write units, where each unit has a logical address. At block 304, a data placement algorithm obtains a physical address for each logical address. The data placement algorithm executes on the host computer 100 software module 106 and determines the physical address based on selected data (e.g., metadata) on the host computer 100, such as wear information and other flash device information. In an embodiment, the physical write location determined by the placement algorithm is selected so as to equalize wear across different physical locations, using determination algorithms well known in the art. In another embodiment, the physical location determined by the placement algorithm is such as to reduce write amplification from future garbage collection operations; the algorithms to determine such a physical location are well-known in the art. In another embodiment the solid-state storage devices include a mixture of devices with different performance characteristics, for example single-level cell (SLC) and multi-level cell (MLC) flash. In this case the placement algorithm takes the physical characteristics of devices into account while determining the physical location. For example, data that is expected to be rewritten frequently is placed on a high-endurance device such as SLC flash, while infrequently updated data is placed on low-endurance devices such as MLC flash. In an embodiment, establishing a correspondence between a logical address and a physical address is responsive to at least one of a measurement of wear of at least one of the solid-state storage devices, projected performance of the main board, and characteristics of at least one of the solid-state storage devices. In an embodiment, establishing a correspondence between a logical address and a physical address is responsive to prior output of an ECC decoder. Thus, information about the location of prior errors is taken into account when deciding on a physical location for the data.
The requests are then queued at an outbound request queue in block 306. In block 308, the hardware module 102 is checked to see if it is CTS the write requests. In block 310, if the module is not cleared, the system will wait until the hardware module 102 indicates it is clear. If the hardware module 102 is CTS, the requests, including the data to be written along with the logical and physical addresses, are sent to the hardware module 102, as shown in block 312. In an embodiment, blocks 300-312 are performed by the host computer 100, and blocks 314-322 are performed by the hardware module 102.
At block 314, the write requests for each of the logical and physical addresses are received. In block 316, the hardware module 102 determines if there are outstanding write requests, for example at the same hardware resource (possible hardware resources include a flash module, a memory device or chip, a die, of plane) where the write operation has to be performed. If there are outstanding write requests, the module waits, as shown in block 317. If there are no outstanding write requests, the hardware module 102 writes to update the FTL at the specified logical address for each physical address, as shown in block 318. At block 320, the write command is sent to the flash memory device or physical address of the memory, where the physical address matches the address written into the FTL in block 318. In block 322, a write completion acknowledgement is sent to the host computer 100.
FIG. 4 is a block diagram of process for a garbage collection operation in accordance with an embodiment. At block 400, the host computer 100 finds the best candidate block of data to erase during the operation. In an embodiment, the best candidate is the block with the least number of valid pages. In another embodiment, the candidate is selected because it has the least amount of wear. The host computer 100 may use metadata and other information stored on the host computer100 that describes flash memory properties to identify the candidate for the erasing operation. Further, the metadata on the host computer 100 also provides the physical addresses of the block of data selected for erasing. Metadata stored at the host includes one or more of the following: valid status of each physical page, status of each block, possibly including current write page, erase status, erase count, either precise or approximate, and current operation being performed if any. In an embodiment this information is processed by an independent thread of execution. In additional embodiments this information is processed when a garbage collection operation is triggered.
At block 402, a read physical address flash operation, such as that described below in FIG. 5, is queued for any valid pages within the block of data identified at block 400. In an embodiment, the physical addresses of the valid pages are determined using an offset value from the starting physical address of the block of data to be erased by the garbage collection process. The physical addresses of the valid pages to be erased are queued, and at block 404, the write of the valid pages and logical addresses returned by the physical read are queued. The logical address associated to a physical page is obtained from metadata appended to the page contents at write time and retrieved alongside with the page contents when performing a read operation. At block 406, the erase requests for the block of data to be erased are queued after all the valid pages have been written (i.e., copied) in block 404 to another location. The process maintains the logical address as the physical addresses of valid pages are changed, where the FTL is updated with the new address for the data. In an embodiment, the processing depicted in FIG. 4 is performed by the host computer 100.
FIG. 5 is a block diagram of a process for a read physical address flash operation according to an embodiment. At block500, a request for reading data from a physical address is queued to the outbound request queue of the host computer100. At block 502, the hardware module 102 is checked to see if it is CTS the read requests. At block 506, if the module is not cleared, the system will wait until the hardware module 102 indicates it is clear. If the hardware module 102 is cleared to send, the read requests and corresponding physical addresses are sent to the hardware module 102, as shown in block504. In an embodiment, blocks 500-504 are performed by the host computer 100, and blocks 508-514 are performed by the hardware module 102.
In block 508, the read requests for each of the physical addresses are received. At block 510, the hardware module 102determines if there is an outstanding physical address read request. If there are outstanding read requests, the module waits, as shown in block 511. If there are no outstanding requests, the read command is sent to the specified physical address in block 512. After the data has been read at the specified physical address, the data and the logical address are delivered to the host computer 100 by the hardware module 102. The read physical address operation may be performed when the host computer 100 knows the physical address of the data it is requesting, such as for a garbage collection operation as described above in FIG. 4.
In additional exemplary embodiments, only an acknowledgment is provided to the host computer 100 after the data is read. The read data is stored on the hardware module 102, for example in a buffer on the controller sub-module 108 or in other memory available at the hardware module 102. In additional embodiments the logical address associated with the page read is communicated back to the host computer 100. In additional embodiments the logical address is stored in a buffer available at the hardware module 102. Whenever the data is stored in a buffer at the hardware module 102, the write operation required to save the page to be erased does not include transferring the page data from the host computer 100, but only requesting that the data used in the write operation is obtained from the buffer at hardware module 102.
FIG. 6 is a block diagram of a process for a flash erase operation according to an embodiment. At block 600, a physical address of the block of data in the flash to be erased is obtained. At block 602, the erase request is sent to the outbound request queue. At block 604, the hardware module 102 is checked to see if it is CTS the erase request. At block 608, if the module is not cleared, the system will wait until the hardware module 102 indicates it is clear. If the hardware module 102 is cleared to send, the erase requests and corresponding physical addresses are sent to the hardware module 102, as shown in block 610. In an embodiment, blocks 600, 602, 604, 608 and 610 are all performed by the host computer 100, and blocks 612, 614, 616 and 618 are performed by the hardware module 102.
At block 612, the erase requests for each of the physical addresses are received. At block 614, the hardware module 102determines if there is an outstanding erase requests. If there are no outstanding erase requests, the module waits, as shown in block 615. If there is an outstanding request, the erase command is sent to the specified physical address and flash memory device 118 at block 616. After the data has been erased at the specified physical address, an acknowledgement of the erased data is communicated to the host computer 100 by the hardware module 102. The erase operation may be performed when the host computer knows the physical address of the data it is requesting to erase, such as for a garbage collection operation as described above in FIG. 4.
Embodiments provide a high performance solid-state device management solution that can be used with low cost volume processors. Specifically, embodiments include a split of flash management complexity between a software module running on a host processor and a hardware module that includes the flash memory. In embodiments, the host processor and system are located on a main board of a computer system which the hardware module is a card that includes the flash memory, a controller sub-module and a fast memory module. The fast memory module (including the FTL) on the hardware module hosts information used for logical to physical address translation (e.g., a table) to provide fast read operation performance. By placing the address translation information on the hardware module, valuable memory on the host is saved. Other metadata describing the flash memory is located on the host to utilize the host processor. Operations performed on the host using the metadata on the host utilize the host processor and do not negatively impact the performance of the hardware controller connected to the flash. The division of data and operations between the host and hardware module also simplifies the system while reducing cost by not having a processor on the module.