SYSTEM AND METHOD FOR MULTISTAGE PROCESSING IN A MEMORY STORAGE SUBSYSTEM

Abstract

Embodiments of this disclosure relate to improving solid-state non-volatile memory management. Embodiments improve the management of solid-state non-volatile memory by providing an execution manager responsible for controlling the timing of providing a request to a memory unit for execution. In embodiments, the execution manager traverses a list of received requests for memory access and dispatches commands for execution. In embodiments, if a request is directed to memory units which have reached a threshold for outstanding requests, the request may be skipped so that other requests can be dispatched for memory units which have not yet reached the threshold.

Description

BACKGROUND OF THE INVENTION

Systems for managing access to solid-state memory often perform complicated tasks such as maintaining drive coherency and mapping logical to physical addresses. In addition to receiving and processing memory access requests from host systems, these management systems often also process memory access requests related to internal maintenance operations such as garbage collection and wear leveling. The management systems may also be configured to respond to changing conditions in the physical memory media, including handling return status of memory commands executed in the media.

BRIEF DESCRIPTION OF THE DRAWINGS

Systems and methods which embody the various features of the invention will now be described with reference to the following drawings, in which:

FIG. 1 illustrates an overview of an embodiment of a media access control system.

FIG. 2 illustrates a more detailed overview of an embodiment of a media access control system.

FIG. 3 is a flowchart showing the processing of a media access request object according to an embodiment.

FIG. 4 illustrates a structure of an embodiment for a media access request object.

FIG. 5 illustrates a detailed overview of an embodiment for a media access request manager.

FIG. 6 illustrates a detailed overview of an embodiment for a media access request executor.

FIGS. 7A-7D illustrate the handling media request objects by a media access request executor in accordance with an embodiment.

DETAILED DESCRIPTION

While certain embodiments of the inventions will be described, these embodiments are presented by way of example only, and are not intended to limit the scope of the inventions. Indeed, the novel methods and systems described herein may be embodied in a variety of other forms. Furthermore, various omissions, substitutions and changes in the form of the methods and systems described herein may be made without departing from the spirit of the inventions. The accompanying claims and their equivalents are intended to cover such forms or modifications as would fall within the scope and spirit of the inventions.

Overview

Embodiments of the invention described herein are directed to pipelined or multi-staged systems and methods for controlling and managing access to physical memory media in solid-state storage subsystems. In an embodiment, the system separates media access request management tasks such as address translation, preparation for execution at the physical memory media, and completion status reporting into discrete stages and processing components to improve performance. In an embodiment, the components in the various stages of processing media access requests cooperate with one another by using a common data structure termed a media access request object to share information on a pending request that is being processed and report any result data and/or completion status information to an entity that originated the media access request. In an embodiment, the components in the various stages of processing media access requests are executed on different processors and/or processor threads. In an embodiment, the system manages the execution timing of requests on memory units (e.g., dies) of a solid-state storage subsystem to improve concurrent use of the memory units.

An overview of a solid state storage subsystem 120 implementing a multi-stage media access control system embodiment is presented in FIG. 1. As shown in FIG. 1, a plurality of media access requesters 101 within the storage subsystem 120 may have requests for memory access to memory arrays 105 in the storage subsystem 120. In an embodiment, a host device 100 is in communication with at least one of the media access requesters 101. The host device 100 may have requests to read or write data stored in memory arrays 105. In an embodiment, the plurality of media access requesters 101 include a media access requester responsible for handling host commands received from the host device 100, a media access requester responsible for issuing garbage collection commands, and a media access requester responsible for issuing wear leveling commands.

As shown in this embodiment, the plurality of media requesters 101 send media access requests to a media access request manager 102 in the storage subsystem 120. In an embodiment, the media access request manager 102 performs any necessary processing including address translation and ensures the media requests can be executed on the memory arrays 105. The address translation performed by the media access request manager 102 can include translation from a logical to a physical address. In an embodiment, once translation is complete, the media access request manager 102 then sends the media access requests to a media access request executor 103 in the storage subsystem 120. The media access request executor 103 is in communication with a media controller 104, and in an embodiment the media access request executor 103 determines when to send media requests to the media controller 104 for execution in the memory arrays 105. In an embodiment, since the media access requests are embedded in media access request objects, the media access request executor, as part of the sending process, may also perform any required translation of data within the media access request objects to data in a format that is understandable by the media controller 104. The media controller 104 performs media requests on the memory arrays 105 and returns a status and/or an error message to the media access request executor 103. In an embodiment, the media access request executor 103 in turn returns the status/error message of media access requests to the plurality of media access requesters 101.

Storage Subsystem Overview with Media Request Objects and Pointers

FIG. 2 provides an additional overview of the storage subsystem illustrating the use of media access request objects and pointers to process media access requests according to an embodiment. As shown in FIG. 2, a plurality of media access requesters 201 may comprise several requesters, such a host interface requester 201a, a garbage collection requester 201b, and a wear leveling requester 201c. In this embodiment, a host 200 communicates with the host interface requester 201a, but in other embodiments the host may communicate with a plurality of requesters. When the media access requesters 201 want to perform a memory access command on and interface with memory arrays 208, the media access requesters first obtain a media access request pointer to a media access request object from a free pointer pool 202a associated with a shared pointer memory 202. In this embodiment, the shared pointer memory 202 is stored in Static Random Access Memory (SRAM), though storage in other types of memory is possible in other embodiments. The media access requesters 201 may each have a dedicated portion of the shared pointer memory 202 associated with each media access requester, or the media access requesters may each request a pointer from a common pool of pointers.

In an embodiment, the shared pointer memory 202 includes a memory structure 202b with memory allocated to memory access request objects. The shared pointer memory 202 can also include a plurality of pointers to the memory access request objects. In an embodiment, each media access requester is associated with a dedicated pool of memory and may maintain its own free pointer pool.

As described below, in an embodiment, each component may access and update information in a memory access request object located in shared pointer memory 202. Each component's access is denoted by dotted lines in FIG. 2. In an embodiment, each component may be executed on a separate thread on a processor. In an embodiment, one or more components may be located on a different processor. Though in these embodiments the components are executed on different processors or separately executing threads, in some embodiments the components nonetheless all access shared pointer memory 202. Contention between processing units for access to shared pointer memory 202 is controlled by logically programming each component to access only memory associated with memory access request pointers currently being handled by that component. For example, once a media access requester 201 completes its processing of a media access request object, it passes the associated pointer to the media access request manager 203 and thus relinquishes to the media access request manager 203 any access right to that object. Likewise, when the media access request manager 203 completes processing of the same object and passes the associated pointer to the media access request executor 206, the media access request manager 203 no longer has access to the object. In summary, in various embodiments, the processing unit that has the pointer is the sole unit with access to the referenced object. Data integrity in the shared memory objects is preserved even as they are accessed by different processing units executed on different processors and/or threads, without the use of a synchronization and/or locking mechanism that may require substantial implementation costs and/or run-time system resources.

Returning to the top of FIG. 2, after obtaining a media access request pointer, the media access requester can populate the referenced media access request object with information relating to the media access request to be issued by the media access requester. For example, the media access requester may populate the media access request object with request data such as a logical address, data indicating the type of memory operation to be performed (e.g., read, write, or erase), and data associated with the operation (e.g., data to be written to a memory array for a write operation). Additional details on request data will be provided below in conjunction with the description of FIG. 4.

Once the media access request object is populated with request data, the media access requesters 201 can then pass the pointer for the media access request object to the media access request manager 203. The passing of the pointer, rather than the object itself, reduces the memory footprint necessary to implement this shared message architecture and improves system performance. In an embodiment, the media access request manager 203 maintains an invalid page table 205 and a mapping table 204. The mapping table 204 maintains a mapping between logical and physical addresses, while the invalid page table 205 maintains a list of physical addresses that contain invalid data. Typically, addresses (e.g., pages) referenced by the invalid page table can be freed through a garbage collection operation and made ready for further use. In an embodiment, the media access request manager 203 processes the incoming requests and resolves any contention between requests submitted by the plurality of media access requesters 201. For example, if the execution of a request requires an alteration to the mapping table 204 and/or invalid page table 205, the media access request manager 203 may update the tables to reflect the alteration. In certain circumstances, a media request may specify a media operation on data that has been altered since the request was first issued by the originating media access requester. In these circumstances, the media access request manager 203 may alter the request to conform to the alterations, or the media access request manager 203 may reject the request due to the conflict and return the associated media access request pointer to the originating media access requester.

In an embodiment, the media access request manager 203 may also populate the physical address field of the media access request object with a physical address that corresponds to a logical address indicated in the media access request object. For a read operation, determining a physical address that corresponds to a logical address may involve a simple lookup in the mapping table 204. For write operations, the media access request manager 203 may maintain a list of upcoming physical write locations associated with the media access requester that submitted the request. In an embodiment, the upcoming physical write locations may be associated with an open group of blocks, and the media access request manager 203 can choose a write location from the open group of blocks that is currently assigned to the requester. In an embodiment, the media access request manager 203 is configured to assign one or more open groups of blocks to individual requesters 201. In further embodiments, other ways of assigning a write location to a logical address may be implemented in any way understood to one skilled in the art.

In an embodiment, after processing the request as described above, the media access request manager 203 passes the media access request pointer to a media access request executor 206. The media access request executor 206 may be in communication with a media controller 207 to execute commands on memory arrays 208. The media access executor 206 schedules execution of requests on the memory arrays 208. In an embodiment, each memory array 208 has an associated execution queue, which stores commands issued by the media controller 207 in response to requests dispatched by the media access executor 206. In an embodiment, the media access request executor 206 maintains a record of the outstanding commands in each execution queue, and passes requests to the media controller according to the outstanding commands in each execution queue. In some embodiments, the media access request executor 206 passes the media access request pointer to the media controller 207, and the media controller 207 executes a command specified by the referenced media access request object on the memory arrays 208. In other embodiments, the media access request executor 206 translates data within the media access request object into a format executable by the media controller 207 on the memory arrays 208.

After execution of the media request on the memory arrays 208, the memory arrays 208 and/or media controller 207 in an embodiment return result data such as a status (e.g., indicating success or failure) and any other information associated with the execution to the media access request executor 206. For example, associated information for a read operation may include the data read from the memory arrays. In certain embodiments, the media access request executor 206 populates a status field in the media access request object with the returned result data and other returned information. In other embodiments, the media controller 207 populates the media request object data fields with result data.

Once result data is inserted into the media request object, in an embodiment the associated media access request pointer can then be passed from the media access request executor 206 to a request status pool/status reporter 209. In an embodiment, the media access requesters 201 will poll the request status reporter/pool 209 to retrieve any updated status for requests referenced by media access request pointers associated with that requester. In another embodiment, the request status pool/reporter can push status information to the associated requester(s).

Media Access Request Processing Cycle

FIG. 3 is a flowchart depicting the processing cycle of a media access request pointer/object according to an embodiment. As shown in this embodiment, at block 300, a media access request pointer is maintained in a pool of available request pointers. When a media access request is needed by a media access requester, a media access request pointer can be allocated to the requester at block 301. Next, as shown at block 302, the media access requester can populate the media access request object referenced by the allocated pointer with command data, such as the command type, the logical address associated with the command, and any other associated data (such as write data for a write command). The media access request pointer is then passed to the media access request manager at block 303, where the media access request data may be updated for translation. If the media access request is determined to have a conflict with other pending requests or otherwise cannot be completed according to the data populated in the referenced media access request object, the media access request may be updated with a status indicating it was not executed and the media access request pointer may be passed to a status pool 306. After updating the media access request object with a physical address location by translation of the logical address, the media access request manager can pass the media access request pointer to the media access request executor at block 304. The media access request is then executed on the media at block 305 and the request will be updated with a result and status. As described above, the media access request pointer may or may not require translation for a media controller for execution at block 305. After execution, the media access request pointer is passed to the status pool 306. As described above, the status pool 306 holds media access request pointers until they are reviewed by the media access requesters. Once reviewed, the media access requester will then free the pointers by placing them back into the pool of available request pointers, where they can be allocated to the next requesters.

Media Access Request Object Structure

FIG. 4 depicts data fields allocated for the media access request object according to an embodiment. As shown in the embodiment in FIG. 4, the media access request object includes a variety of data fields to hold information about the request. The media access request object embodiment shown in FIG. 4 includes data fields for a macro pointer 401, a single/dual plane designation 403, plane selection 405 indicating which plane on a die is being accessed, operation data 408 indicating the associated memory operation type, a logical chunk number 402, a physical chunk number 404, a requester ID 406, a result status 407, and a count 409 for the number of buffers in the request. The media access request object may also contain a field for other data 410. In certain embodiments, the logical chunk number data field 402 may also be used for metadata about the request. Certain embodiments of the media access request object may include data for specifying single or dual plane operations which can be designated by the single/dual plane designation 403. The media access request object may also include buffers 411 to store read and write data relating to the request.

As the media access request object is passed among the components of the media request control system within the storage subsystem, various portions of the media access request object are populated. For example, the media access requester may populate fields such as the requester ID data field 406 (indicating the originating requester), the operation data field 408, and the logical chunk number data field 402. For a write request, the media access requester may populate data to be written into buffers 411. In an embodiment, when the media access request object is passed to the media access request manager, the physical chunk number field 404, and the plane field 405 are populated. In an embodiment, if the request cannot be executed, the media access request manager can update the result data field 407. After the execution of the media request in the physical memory media, the media access executor and/or the media controller can update the result data field 408, and for some operations, e.g., a read operation, the buffers 411 with data read from the target physical address.

Media Access Request Manager

A detailed view of an embodiment of the media access request manager is shown in FIG. 5. In this embodiment, three media access requesters are shown, a garbage collector 503, a user data manager 502, and a requester for other internal operations 501. Other internal operations 501 may include, for example, wear leveling operations. The user data manager 502 is in communication with a host interface 500 and is responsible for responding to host communications (e.g., accepting incoming memory commands from the host). In this embodiment, each requester may have a dedicated request queue 504a-c. In other embodiments, certain requesters may share one or more request queues 504. For example, the user data manager 502 and other internal operations 501 may share a single request queue. Each media access requester may obtain a media access request pointer and populate the media access request object as described above. In an embodiment, the media access requester can then place the media access request pointer in the associated request queue.

In an embodiment, the media access request manager 505 receives media access request pointers from the request queues 504a-c. After receiving the media access request pointers, the media access request manager 505 in an embodiment updates the media access request object with reference to a mapping table 506 and an invalid page table 507. In an embodiment, the media access request manager 505 maintains a designated write location for each requester or request queue. In an embodiment, the media access request manager 505 determines a physical address for the write location corresponding to an open block in a group of blocks for each requester or request queue. In an embodiment, after receiving a write command request at the media access request manager 505, the mapping table 506 is updated with the new physical address corresponding to the logical address of the request. The invalid page table 507 is then updated to indicate that the physical address formerly mapped to the logical address is invalid. In an embodiment, the media access request manager 505 also updates an invalid page counter for the block or group of blocks corresponding to the old physical address. In an embodiment, the media access request manager 505 is the only entity with access rights to write to the mapping table 506 and the invalid page table 507. Restricting write accesses to the media access request manager 505 ensures a single point of control for these tables, and prevents possible contentions and data coherency errors that may result if the requesters were granted direct access to the tables.

In an embodiment, when the media access request manager 505 receives media access request pointers from the request queues 504a-c, the media access request manager 505 may determine a ratio at which to process media access request pointers from each queue. While requests from each requester must generally be executed in-order relative to other requests from that requester, requests from each requester may generally be executed in a different order than received relative to requests from another requester. For example, if a read command from the user data manager 502 is directed to the same logical address as a garbage collection command from the garbage collector 503, the media access request manager 505 can ensure that the physical address of each command is updated to appropriately compensate for the order of execution. If the garbage collection command is processed first, the read command can be directed to the new physical address corresponding to the same logical address. If the read command is processed first, the read command is executed prior to any data movement associated with the garbage collection command and revisions to the physical addresses of the data. As such, the media access request manager ensures coherency with the mapping and invalid page tables, thereby preventing downstream request execution conflicts.

In an embodiment, the option of processing requests from the queues in different ratios allows the media access request manager 505 to throttle, or provide dynamic ratios, of requests received from each request queue 504 and thereby prioritize a specific requester over other requesters. For example, the media access request manager 505 can receive and process two or three or more requests from the user data manager 502 for every request from garbage collector 503 (or vice versa). In an embodiment, the media access request manager 505 temporarily halts receiving media access request pointers from a request queue when certain conditions are met. For example, if the invalid page table shows a high need for garbage collection, the requests for user data manager 502 may be throttled down or halted to allow for more garbage collection requests to be processed. Alternatively, if the user data manager 502 has an unusually high number of requests, or if the request queue 504a for the user data manager 502 is nearing a threshold, the requests associated with the garbage collector 503 may be throttled down or halted. After preparing the media access request for execution, the media access request manager 505 passes the media access request pointer to the media access request executor.

Media Access Request Executor

FIG. 6 depicts a media access request executor according to an embodiment. In an embodiment, the media access request executor 602 receives media access request pointers from a media access request manager 601. The media access request executor improves throughput and concurrency on memory arrays 607 by determining when to pass media access requests down to a media controller 605 for execution. In an embodiment, the media access request executor 602 maintains a request list 603 and execution counters 604. In this embodiment, received media access requests from the media access request manager 601 are placed on the request list 603. In certain embodiments, the request list identifies a specific memory unit associated with the media access request. The memory unit associated with a media access request may be a die of solid-state memory. In other embodiments, the request list identifies another granularity of the memory arrays 607 (e.g., plane(s) or block(s) with a die or a device including multiple dies), which is generally the lowest level of memory access on which the memory array can accept commands while other commands can be executed on a portion of the memory array addressed at the same level. In this embodiment, the execution counters 604 maintain a record of the pending requests associated with each memory unit. As requests are directed to each memory unit and requests are completed at each memory unit, the execution counter of each memory unit increments and decrements accordingly.

In an embodiment, when the media access request executor 602 has determined that a media access request should be executed, the media access request executor 602 will pass the media access request to media controller 605. As described above, in certain embodiments the media access request executor 602 can pass a media access request pointer to media controller 605, while in other embodiments the media access request executor 602 performs a translation to a format acceptable to the media controller 605. Media controller 605 directs each media access request to an execution queue 606 associated with each memory arrays 607. Each memory array 607 may be accessed via an associated data channel. The execution queue 606 for each memory array 607 may be maintained on the memory arrays 607, or may be maintained as a part of media controller 605. As media access requests are completed on memory arrays 607, the media controller 605 will return a status and any associated information to the media access request executor 602. As necessary, the media access request executor 602 incorporates the status and associated information into the media access request object and passes the media access request pointer to status pool 608. In an embodiment, the media controller 605 populates the status and associated information fields of the media access request object instead of the media access request executor 602. As described above, the media access requesters receive media access request pointers from the status pool 608.

One example method of determining when to execute media access requests is shown in FIGS. 7A-7D. As shown in FIG. 7A, in an embodiment the request list includes a reference to the memory unit associated with each memory access request (each box represents a memory access request and the number within is the reference). In this example, starting with FIG. 7A, the execution counters show that all of the memory units (0-7) do not have any pending media access requests. In an embodiment, the media access request executor also maintains a current position on the request list, which is shown in this example by the arrow pointing to the first request directed at memory unit 1. As the media access request executor sends requests to the media controller for execution, the media access request executor updates the execution counter corresponding to the memory unit associated with the media access request. FIG. 7B shows the status of the request list and execution counters after the first three requests have been processed (now shaded in grey) and sent to the media controller for execution. As shown, two requests for memory unit 1 and one request for memory unit 2 have been sent to the media controller for execution. Accordingly, the media execution counters have been updated, and the request list indicates that these requests have been sent. The current position on the request list as shown is from a time immediately after the last-sent request.

In various embodiments, each memory unit is capable of having a certain threshold number of outstanding requests at once. As such, the media access request executor cannot send additional requests past this outstanding request threshold. The outstanding request threshold is typically dictated by the depth of the execution queue 606 depicted in FIG. 6. In the example shown in FIGS. 7A-7D, for ease of explanation, the threshold number of outstanding requests is two. In actual systems, the number of possible outstanding requests may be higher. In an embodiment, when the current position in the request list points to a request directed to a memory unit for which the outstanding request threshold has been reached, the media access request executor skips that request on the request list and continues to process the next requests on the list. As shown in FIG. 7C, a media access request directed to memory unit 1 has been skipped because that memory unit had already reached the threshold number of outstanding requests (as indicated by the counter). After the skip, the media access request executor has gone on to process requests directed to memory units 3, 4, 2, and 5, and the associated media execution counters have been incremented.

In an embodiment, as the request list is traversed, the current position in the request list will reach a reset point and return to the first skipped request to determine if that request can now be executed. The reset point can be determined in a variety of ways. The current position can return to the first skipped request upon the triggering of a condition. As examples, the condition can be based at least partially on any of: the number of requests that have been skipped (in total or for a particular memory unit), receiving an indication notifying of a request completion from a memory unit that has been skipped, and reaching the end of the request list. In various embodiments, as the request list is continually changing to reflect processed requests and new incoming requests from the media access request manager, periodic snapshots of the request list are taken and the aforementioned condition for reset may also be based on reaching the end of a snapshot of the request list. FIG. 7D shows an embodiment of the request list and execution counters if a reset point has been reached immediately after the system state in FIG. 7C. When the reset point is reached, the media controller is polled to determine if any requests have completed, since these completed requests can result in the media access request executor decrementing the execution counters. In FIG. 7D, requests corresponding to memory units 1 and 2 have completed and the corresponding execution counters are decremented. The current position in the request list now returns to the first skipped request, which here corresponds to memory unit 1. Returning to the first skipped request ensures that memory requests for a particular memory unit are executed in-order relative to that memory unit. Because certain requests are skipped, the requests for a memory unit may be executed out-of-order with respect to the time when the requests for other memory units were received. However, because these requests correspond to different, independently accessible memory units, data coherency is maintained. In the embodiment shown in FIG. 7D, at the reset point the media access request executor also clears the request list of requests that have already been sent to the media controller and/or added to the appropriate execution queues for the respective memory units. Other embodiments may remove a request from the request list only when the request is completed on the memory unit. Completed requests may be passed to a status pool as described above.

In alternate embodiments, other methods for maintaining a list of requests to be executed may be used. For example, at the media access request executor level, the received requests may be placed in individual queues associated with each memory unit. The media access request executor may then send media access requests from the queues to the media controller by checking whether the associated media unit has reached a threshold in the media execution counter. In an embodiment, the media access request executor may check only those queues which have not reached the threshold in the execution counter. In an embodiment, the media access request executor checks the queues associated with each memory unit by a round robin algorithm. A flag can be used to designate which memory units have an execution counter that has reached the threshold. According to another embodiment, the media access request executor reorders the media access requests in the request list at least partly based on the status of the associated memory units. The status of the associated memory units may be determined with reference to the execution counters. Other data management means can be used to control the flow of media requests to the media controller in accordance with this disclosure.

By temporarily skipping certain requests, the media access request executor increases the concurrent use of the memory units (by sending requests to those units before other units have finished requests). Further, by using a reset point, the media access request executor can return to process skipped requests to preserve data coherency.

CONCLUSION

The features and attributes of the specific embodiments disclosed above may be combined in different ways to form additional embodiments, all of which fall within the scope of the present disclosure. Although the present disclosure provides certain preferred embodiments and applications, other embodiments that are apparent to those of ordinary skill in the art, including embodiments which do not provide all of the features and advantages set forth herein, are also within the scope of this disclosure. Accordingly, the scope of the present disclosure is intended to be defined only by reference to the appended claims.

Claims

1. A storage subsystem, comprising: a controller;a non-volatile solid-state memory array, the non-volatile solid-state memory array including a plurality of memory units in electrical contact with the controller;wherein the controller is configured to:receive a plurality of requests for access to the non-volatile solid-state memory array, process each received request for execution and associate each received request with a memory unit in the plurality of memory units;maintain a list of the plurality of received requests;maintain a record of a number of outstanding requests associated with each of the plurality of memory units;selectively send a request from the list of received requests to the memory unit associated with the request;wherein a request is selected based at least in part on determining if a number of outstanding requests associated with a memory unit associated with the request is less than an outstanding request threshold.

2. The storage subsystem of claim 1, wherein the controller is further configured to selectively send a request by: traversing the list of the plurality of received requests,skipping a request if it is determined that the number of outstanding requests associated with a memory unit associated with the request is more than an outstanding request threshold, andreturning to a reset traversal point on the list to re-evaluate the skipped requests after a condition has been reached.

3. The storage subsystem of claim 1, wherein the condition is based at least partially on one of: a number of requests that have been skipped, a number of requests that have been sent, the traversing reaching the end of the list of received requests, the traversing reaching the end of a snapshot of the list of received requests, and a completion of a request associated with a memory unit which has been skipped.

4. The storage subsystem of claim 1, wherein the received plurality of requests for access are not configured for execution on the memory units and the controller is further configured to translate the requests for access to commands executable on the memory units prior to sending to the memory units.

5. The storage subsystem of claim 1, wherein the received plurality of requests for access are configured for execution on the memory units without translation by the controller.

6. The storage subsystem of claim 1, wherein the received plurality of requests for access are configured for execution on the memory units without modification by the controller.

7. The storage subsystem of claim 1, wherein each memory unit comprises a die.

8. The storage subsystem of claim 1, wherein each memory unit is accessible via an associated data channel.

9. A method of executing memory access requests on a plurality of memory units in a storage subsystem, the method comprising: receiving, at a controller, a plurality of requests for memory access;preparing, at the controller, the plurality of requests for execution and associating each request with a memory unit in the plurality of memory units;maintaining, by the controller, a list of the plurality of received requests;maintaining, by the controller, a record of a number of outstanding requests associated with each of the plurality of memory units; andselectively sending, by the controller, a request from the list of requests to the memory unit associated with the requests, wherein a request is selected based at least in part on determining if a number of outstanding requests associated with a memory unit associated with the request is less than an outstanding request threshold.

10. The method of claim 9, further comprising: traversing, by the controller, the list of the plurality of requests;skipping, by the controller, a request if the number of outstanding requests associated with a memory unit associated with the request is more than an outstanding request threshold; andreturning, by the controller, to a traversal point on the list to re-evaluate the skipped requests after a condition has been reached.

11. The method of claim 9, wherein preparing the plurality of requests includes translating the requests for execution on the plurality of memory units in the storage subsystem.

12. The method of claim 9, wherein the condition is based at least partially on one of: a number of requests that have been skipped, a number of requests that have been sent, the traversing reaching the end of the list of received requests, the traversing reaching the end of snapshot of the list of received requests, and a completion of a request associated with memory unit which has been skipped.