DE102021117739B4 - Management of multiple applications' access to a data store - Google Patents

Abstract

Method (200) for managing accesses from a plurality of applications (15n) to a data memory (20) via time frames (TR) with predetermined time periods (td), the plurality of applications (15n) each having at least one queue group (17n) during their execution generate at least one queue (18nm) for listing access commands (19) to the data memory (20), the method having:obtaining (210) individual access volume quotas (Cn), each of which is allocated to one of the plurality of queue groups (17n), wherein the sum of the access volume quotas can be fully utilized within a predetermined time period (td) of a pending time frame (TR);starting (220) the time frame (TR) with the predetermined time period (td);performing (230) a first arbitration phase ( A1), in which, in a single round of administration, at least one of the listed access commands (19) is permitted in the queue groups (17n) of all applications (15n) in succession in all queues (18nm), regardless of a remaining duration of the time frame (TR), if the access volume quota (Cn) allocated to the respective queue group (17n) has not yet been used up; andcarrying out (240) a second arbitration phase (A2), in which in several management rounds in succession in the queue groups (17n) of all applications (15n) in each case at most one of the listed access commands (19) is permitted if the respective queue group (17n) allocated access volume quota (Cn) has not yet been used up, the second arbitration phase (A2) being terminated when the predetermined period of time (td) of the time frame has elapsed.

Description

The present invention relates to computer systems in which multiple applications access a data memory during their execution.

In particular for deterministic / safety-critical systems, the applications are usually allocated bandwidths for accessing the data memory. As a rule, access controllers are used in the computer systems, which manage the applications' access to the data memory, taking into account the bandwidth allocation, as described, for example, by R. Wijayaratne et al. described in the article "Providing QOS guarantees for disk I/O" on pages 57-68 in Multimedia Systems, Vol. 8, 2000. With conventional access controllers, however, the bandwidth allocations between the applications can often not be fully maintained or utilized in practice. And U.S. 2003/0128707 A1 discloses an analog management system for variable length frame transmissions from multiple output queues with different transmission priorities.

It is the object of the invention to provide improved management of accesses by a number of applications to a data memory, which allows bandwidth allocations between the applications to be taken into account more effectively.

This object is achieved by the method for managing accesses by a number of applications to a data memory with the features of claim 1. Particularly advantageous refinements and developments of the invention are the subject matter of the dependent claims.

The method for managing accesses by multiple applications to a data memory over time frames with predetermined time durations, the multiple applications generating at least one queue group each with at least one queue for listing access commands to the data memory during their execution, has the following steps. First, individual access volume quotas each allocated to one of the plurality of queue groups are obtained, wherein the sum of the access volume quotas can be fully consumed within a predetermined period of time of an upcoming time frame. Then the time frame is started with the predetermined period of time. In the time frame, a first arbitration phase is performed first and then a second arbitration phase is performed. In the first arbitration phase, at least one of the listed access commands is permitted in a single administration round in the queue groups of all applications, one after the other in all queues, regardless of a remaining period of time in the time frame, if the access volume quota allocated to the respective queue group has not yet been used up. In the second arbitration phase, at most one of the listed access commands is permitted in the queue groups of all applications in several consecutive administration rounds, if the access volume quota allocated to the respective queue group has not yet been used up, the second arbitration phase being terminated when the predetermined period of time of the time frame has expired.

The individual access volume quotas are each allocated to a queue group, and the applications each create at least one queue group, each with at least one queue, so that the access volume quotas are allocated to a queue group of an application, but not to individual queues of an application. The individual access volume quotas are reallocated to the queue groups for each time frame, so they can be different for different time frames depending on the access needs and depending on the length of the time frame.

The formulation in the first arbitration phase that at least one of the listed access commands is allowed in all queues of all queue groups naturally also includes not allowing any access commands in a queue if the respective queue contains no access command. This formulation in the first arbitration phase, that at least one of the listed access commands is allowed in all queues of the queue groups, also includes allowing more than one access command, for example by multiple sequential management of the queues when managing the respective queue group in the respective management round and / or in the Case of access commands of several different types of access in the queue by allowing one access command from each of the types of access present in the queue in a management of the queue. The access types are explained in more detail below.

The formulations in the second arbitration phase, that in several successive administration rounds in the queue groups of all applications at most one of the listed access commands is permitted and the second arbitration phase is ended when the predetermined time period of the time frame has expired, includes depending on the time situation in particular also cases in which the queue groups of all applications are resolved several times (multiple management rounds), the queue groups of all applications are only resolved once each (only a complete management round), or a part of the queue groups is only resolved once (only a partial management round). This formulation in the second arbitration phase, that in several administration rounds in succession in the queue groups of all applications at most one of the listed access commands is allowed, of course also includes not allowing any access commands in a queue group if none of the queues of the respective queue group contains an access command. Allowing at most one access command in a queue group can be achieved in particular by the queues of a queue group only being managed one after the other until an access command that can be allowed is recognized in the respective queue group.

For a better understanding of the two different arbitration phases of the subject matter of the invention as defined in the appended claims, these will now be described again by way of example and in greater detail. In the first arbitration phase, access commands are permitted for processing in each queue group in a group round (pass through all queue groups) until there are no more access commands in the queues for which the associated access volume quota is available. Within each queue group, only one access command of the respective queue is permitted for processing over several queue rounds (cyclic access to the queues of a queue group). In the first arbitration phase, several access commands can be allowed for processing per queue. Allowing access commands to be processed ends in each queue when the access volume quota associated with the next access command contained in the queue has been used up or the queue is empty. As soon as one of these two conditions is given for all queues of a queue group, the arbitration is continued within the first arbitration phase with the next queue group. It should also be noted that the status of the queues can change at any time, since the applications can write new access commands to the queues at any time, which are then read and managed by the access controller. In the second arbitration phase, the queue groups are searched for access commands in the queues until the time frame has ended or all access volume quotas of all queue groups have been used up. The access commands are processed sequentially, with only a maximum of one access command being processed in each group round per queue group. Management of a queue group ends as soon as an access command is found that can be admitted (and possibly also fragmented) or when all queues in a queue round have been searched through without an access command being admitted, with only a single pass through at most all queues within a queue group. If an access command of a queue could be allowed for processing, then the management of the access commands is continued in the next queue of the relevant queue group if the queue group is searched for an access command again after a group round.

Access management in the second arbitration phase preferably begins only after the completion of the processing of all access commands from the first arbitration phase that are permitted for processing, since this way it can be better ensured that the access commands that are sequentially permitted in the second arbitration phase can be processed within the remaining time of the time frame. The processing of the access commands from the queue group permitted in the first arbitration phase may, as far as possible, take place in parallel. In contrast, the processing of the access commands permitted in the second arbitration phase should take place sequentially in order to be able to consider and take into account the remaining time of the predetermined time period of the time frame. The application of the management method is not limited to any specific type of data storage; The proposed management method is particularly advantageous for data memories with a high storage capacity (in particular in the range of several or even several hundred gigabytes).

Carrying out the two special different arbitration phases in the time frame has the advantage that the available bandwidth of the data memory is used very efficiently. What is achieved in particular is that each queue group within a time frame can use the access volume quota assigned to it to the maximum extent on the one hand and as completely as possible on the other. An equal distribution among the queue groups (in the second arbitration phase) and the queues within the queue groups (in the first and the second arbitration phase), this equal distribution being carried out via the continued administration of the queue groups and their queues over a number of time frames, each with the two arbitration phases. This equal distribution is also achieved even though the time frame in which the applications are executed (eg under management of the host system) and the time frame on which the access controller works do not necessarily have to be of the same length or synchronous to one another. In general, the access commands can even be written asynchronously into the queues of the queue groups by the applications (ie also during the two arbitration phases) and there can be more access commands in the queues than can be processed in one time frame.

The queues of the queue groups are preferably each configured to list access commands of a first access type and/or access commands of a second access type. In this case, the access volume quotas allocated to the plurality of queue groups preferably each include a first access volume quota for the first access type and/or a second access volume quota for the second access type. The first and second types of access are, for example, read access and write access. In principle, it is also possible to use more than two types of access (read, write, delete, etc.). While in the first arbitration phase in one administration round in the queues of the queue groups one access command of the first access type and one access command of the second access type can be allowed, in the second arbitration phase in the several administration rounds in each of the queue groups at most one single access command of the first or second type of access allowed.

In a preferred embodiment of the invention, if a residual value of the access volume quota allocated to the respective queue group is not sufficient for the complete execution of an access command, the access command is fragmented in such a way that only part of the access volume of the access command can be processed in accordance with the residual value of the allocated access volume quota. This volume-related fragmentation of an access command can take place in both arbitration phases depending on the status situation.

In a preferred embodiment of the invention, if in the second arbitration phase a remaining time in the time frame is not sufficient for the complete execution of an access command, the access command is fragmented in such a way that only part of the access volume of the access command can be processed, the access time of which corresponds to the remaining time in the time frame . Depending on the status situation, this time-related fragmentation of an access command takes place only in the second arbitration phase, while in the first arbitration phase the access commands can be permitted independently of the remaining duration of the time frame.

The time frame can be better utilized as a result of the two aforementioned fragmentations of access commands. A maximum of one fragmentation is preferably carried out for each type of access. An identifier of the fragmented access command is preferably temporarily stored so that the next processing of the queue groups can begin in a next time frame with the admission management of the access command fragmented in the previous time frame. The fragmentation of the access command is optionally continued over a number of time frames until the respective access command has been completely processed. This prioritization of fragmented access commands in the next time frame compared to non-fragmented access commands is an essential part of the fragmentation aspect that is preferably carried out, since this allows only one access command fragmentation per access type per queue group. If the queues can contain access commands for more than one access type and separate access volume quotas are allocated accordingly, no further access commands are preferably managed by a queue (not even access commands from other access types for which an access volume quota is still available and for which no access command is currently being fragmented), as long as fragments of an access command from this queue are being processed, i.e. as long as an access command taken from this queue in fragments cannot be processed completely in the first arbitration phase. The sequential access to the queues is thus stopped for the duration of the processing of an access command that was last removed from this queue and is currently being processed in fragments.

In one embodiment of the invention, the time frame is terminated before the predetermined period of time has elapsed if the access volume quotas of all queue groups have been completely used up. This measure of ending the time frame early allows the next time frame to start earlier, so that the data throughputs for all queue groups and their queues can be increased. Since such a premature termination of the time frame only takes place when the access quotas of all applications have been completely used up, none of the applications is disadvantaged as a result.

In one embodiment of the invention, an identifier of the queue of a queue group that was last processed is buffered in each case and a next management phase of the same queue group begins with that queue after the queue that was last processed. As an alternative or in addition, an identifier of the queue group last dealt with can also be buffered in each case and a next management phase can begin with that queue group after the queue group last dealt with. These measures allow the queue groups or the queues of a queue group to be managed and dealt with more evenly.

In one embodiment of the invention, to determine whether a remaining period of time in the time frame is sufficient for processing an access command, an access time required by the access command is determined based on a table in which the connection between access times and access volumes is listed. This procedure is preferably also used to calculate the fragmentation of the access volume. With this approach, access management can take into account, for example, that different types of access require different access times (write accesses usually take significantly more time than read accesses for the same volume) and access times for volumes are usually non-linear (an access time for, for example 128 kB is usually significantly less than eight times the access time for 16 kB).

In one embodiment of the invention, obtaining the access volume quotas includes retrieving the access volume quotas from host software, which allocates the access volume quotas to the multiple queue groups of the multiple applications, the sum of which can be fully utilized within the predetermined period of time.

The invention also relates to a computer system which has at least one processor; a data store; a plurality of applications, each of which can be executed on the at least one processor and, during their execution, each generate at least one queue group each having at least one queue for listing access commands to the data memory; and an access controller configured to manage the accesses of the multiple applications to the data store according to the method of the invention described above.

With this computer system, the same advantages can be obtained as with the management method described above. With regard to the advantages, definitions of terms and preferred configurations, reference is made to the above explanations in connection with the management method of the invention.

The access controller can be a hardware component (e.g. microcontroller) or can be implemented by software executable on the at least one processor. Depending on the basic structure of the computer system, the access controller can, for example, control a data access element for processing accesses by the applications to the data memory or be integrated into the data access element. The computing system is implemented, for example, by a single computing device or multiple computing devices, or in the form of a computing platform.

The computer system can also have host software that can be executed on the at least one processor and that, when executed, allocates access volume quotas to the multiple queue groups of the multiple applications, the sum of which can be fully utilized within a predetermined time period of an upcoming time frame.

The subject matter of the invention is also a computer-readable storage medium that contains instructions that can be executed by an access controller of a computer system and, when they are executed, allow the access controller to carry out the above-described method of the invention for managing access by a number of applications to a data memory of the computer system. The computer-readable storage medium can basically be any type of storage medium (e.g. RAM, ROM, EEPROM, flash memory, CD-ROM, DVD, etc.).

The invention, which is defined in the appended claims and described in detail below by way of example, is not limited to any special applications, in particular not to special applications or special data storage devices, nor to special fields of application of the computer systems.

• 1 ein Computersystem gemäß einem Ausführungsbeispiel der Erfindung;
• 2 ein Flussdiagramm des Verwaltungsverfahrens der Erfindung; und
• 3 bis 8 mehrere Teilflussdiagramme zu dem Flussdiagramm von 2 gemäß Ausführungsbeispielen der Erfindung.

The above and other features and advantages of the invention will be apparent from the following description of a preferred, non-limiting embodiment with reference to FIG understand the drawing better. It shows, partly schematically:

• 1 a computer system according to an embodiment of the invention;
• 2 a flow chart of the management method of the invention; and
• 3 until 8th several sub-flowcharts to the flowchart of 2 according to embodiments of the invention.

This in 1 Illustrated computer system 10 contains in the basic structure at least one processor 12 and a software system 14, which is executable on the at least one processor 12. The software system 14 has several (in this exemplary embodiment three by way of example) applications 15a, 15b, 15c and, additionally, optionally host software 16. The applications 15n access a data memory 20 of the computer system 10 during their execution. In this exemplary embodiment, the applications 15n access (in particular reading and writing) the data memory 20 via a data access element 22 . The data memory 20 is a mass memory, for example in the form of an integrated hard disk or an inserted memory card.

During their execution, the applications 15n each generate access commands 19, which are listed in queues 18nm. As in 1 As illustrated, each of the applications 15n has one or more queue groups 17n with one or more queues 18nm for the access commands 19. For example, the first application 15a can have a first queue group 17a with three queues 18a1, 18a2, 18a3 for access commands 19, the second application 15b have a second queue group 17b with two queues 18b1, 18b2 for access commands 19 and a third queue group 17c with three queues 18c1, 18c2, 18c3 for access commands 19, and the third application 15c has a fourth queue group 17d with two queues 18d1, 18d2 for access commands 19 have. The access commands 19 listed by the applications 15n can, if required, contain access commands of a first access type (e.g. read access commands 19R) and access commands of a second access type (e.g. write access commands 19W), access commands of other access types also being possible if required. In the queue groups 17n of the applications 15n, the access commands of the different access types can be listed either separately in different queues 18nm or mixed in common queues 18nm.

The data access element 22 processes the access commands 19 listed in the queues 18nm. To manage these data accesses of the applications 15n to the data store 20, the computer system 10 also has an access controller 24 for managing the queue groups 17n (only the queue groups, while it does not know the applications and the applications are only managed by the host software 16). The access controller 24 is, for example, a microcontroller that controls the data access element 22 or a control system integrated into the data access element 22 . As in 1 Illustrated, the computer system 10 has a program memory (computer-readable storage medium) 25, which contains instructions that can be executed by the access controller 24 for carrying out the method for managing the accesses of the applications 15n to the data memory 20.

The computer system 10 can of course also have other components (e.g. displays, input devices, communication interfaces, etc.) that 1 are not shown because they do not have to be present for the implementation of the data access management according to the invention.

An exemplary embodiment of the method to be carried out by the access controller 24 for managing the accesses of the multiple applications 15n to the data memory 20 is described below with reference to the 2 until 8th illustrated flowcharts explained in detail as an example.

2 first shows the main steps of the method 200 for data access management of the applications 15n to the data memory 20, which are carried out by the access controller 24.

In a first step 210, the access controller 24 (in this exemplary embodiment from the host software 16) obtains individual access volume quotas Cn, which the host software 16 assigns to one of the multiple queue groups 17n that are currently set up by the multiple applications 15n for the pending time frame TR have been allocated. In this exemplary embodiment, in which the queues of the applications 15n can optionally contain read access commands 19R and/or write access commands 19W, the queue groups 17n are each allocated individual read access volume quotas CRn and individual write access volume quotas CWn for the pending time frame TR. The access controller 24 thus obtains from the host software 16 a read access volume quota CRa and a write access volume quota CWa for the first queue group 17a of the first application 15a, a read access volume quota CRb and a write access volume quota CWb for the second queue group 17b of the second application 15b, a read access volume quota CRc and a write access volume quota CWc for the third queue group 17c of the second application 15c, and a read access volume quota CRd and a write access volume quota CWd for the fourth queue group 17d of the third application 15c .

The allocation of the access volume quotas Cn (CRn and CWn) is adapted to the bandwidth that can be processed by the data store 20 such that the sum of the access volume quotas Cn of all applications 15n can be fully utilized within a predetermined time period td of an upcoming time frame TR. The durations td of consecutive time frames TR are unchanged, for example, but can be varied as required or depending on the application. The allocation of the individual access volume quotas Cn to the queue groups 17n of the applications 15n is basically variable for consecutive time frames TR, so that the process sequences of the applications 15n can be taken into account for the allocation of the access volume quotas.

After obtaining the access volume quotas Cn in step 210, the access controller 24 starts the pending time frame TR with the predetermined time period td in step 220. In the time frame TR, a first arbitration phase A1 and then in step 240 a second arbitration phase A2 are carried out in step 230.

After completion of the time frame TR or the two arbitration phases A1, A2, the method 200 returns to step 210. I.e. several time frames TR are usually processed, for which individual access volume quotas Cn are allocated to the queue groups 17n set up by the applications 15n and the two arbitration phases A1, A2 for managing the accesses to the data memory 20 are carried out.

Referring to 3 and 4 Step 230 for performing the first arbitration phase A1 will now be explained in more detail.

In the first arbitration phase A1, in step 232, the access volume quotas Cn for a current queue group 17x (for example the first queue group 17a of the first application 15a) are used up in a manner as described in more detail in FIG 4 is illustrated. After completion of this access quota consumption 232 for the current queue group 17x, it is then checked in step 234 whether this access quota consumption 232 has already been dealt with for all queue groups 17n of all applications 15n. If not (N in 234), in step 236 a next queue group 17x+1 (for example the second queue group 17b of the second application 15b) is selected as the current queue group 17x in order to then carry out the access quota consumption 232 for this next queue group 17x+1. When all queue groups 17n have then been dealt with using method step 232 (Y in 234), this first arbitration phase A1 ends.

4 illustrates the method step 232 of the first arbitration phase A1 for a preferred exemplary embodiment of the data access management of the applications 15n, which contains a fragmentation concept. In this exemplary embodiment, method step 232 of the first arbitration phase A1 contains a first phase 310 in which access controller 24 can allow fragmented access commands 19f that may have arisen in a previous time frame TR by limiting an access command 19, as will be explained later with reference to 7 and 8th explained in more detail. The processing of the fragmented access commands from previous time frames is thus prioritized. In this phase 310, a step 311 checks whether a fragmented read access command 19Rf from the previous time frame exists in the current queue group 17x. If this is the case (Y in 311), it is checked in step 312 whether there is a read access volume quota CRn for the current queue group 17x. If this is the case (Y in 312), the access controller 24 allows the fragmented read access command 19Rf in step 313 (as explained later with reference to 7 and 8th explained in more detail). Then, in a step 314, it is checked whether a fragmented write access command 19Wf from the previous time frame exists in the current queue group 17x. If this is the case (Y in 314), it is checked in step 315 whether there is a write access volume quota CWn for the current queue group 17x. If this is the case (Y in 315), the access controller 24 permits the fragmented read access command 19Wf in step 316 (as explained later with reference to 7 and 8th explained in more detail).

After this prioritization phase 310 relating to fragmented access commands 19Rf, 19Wf, in step 320 an identifier of the last processed queue 18z of the current queue group 17x is retrieved, which has been temporarily stored after its processing in the previous time frame. Then in step 322 it is checked whether for the current queue group 17x the allocated read access quota CRn and the allocated write access volume quota CWn have already been completely used up, ie whether there is no access volume quota residue rCn left for the current queue group 17x in this time frame TR. If this is the case (Y in 322), method step 232 is terminated without processing access commands 19.

If the access volume quotas Cn are not yet completely used up (N in 322), in step 323, starting from the retrieved identifier of the last processed queue 18z of this current queue group 17x, the next queue is accessed as the current queue 18xy of the current queue group 17x for processing. First, in step 324, a check is made as to whether access commands 19 are present at all in the current queue 18xy. If the current queue 18xy is empty (N in 324), the processing of the current queue 18xy stops immediately. If the current queue 18xy is not empty (Y in 324), it is checked in step 325 whether a read access command 19R is listed in the current queue 18xy. If present (Y in 325), it is checked in step 326 whether there is still a read access volume quota remainder rCRn (i.e. remainder of the allocated read access volume quota CRn in this time frame) for this queue group 17x. If this is the case (Y in 326) (regardless of the value of the read access volume quota remainder), in step 327 the read access command 19R is permitted and then returned to step 322 and, if possible (N in 322), in step 323 switched to the next queue .

If there is no read access command 19R in the current queue 18xy (N in 325), based on the check result in step 324 that the current queue is not empty (Y in 324), it is assumed that a write access command 19W is in the current queue 18xy is listed. In step 328 it is therefore checked whether there is still a write access volume quota remainder rCWn (i.e. remainder of the allocated write access volume quota CWn in this time frame) for this queue group 17x. If this is the case (Y in 328) (regardless of the value of the write access volume quota remainder), in step 329 the write access command 19W is allowed and then returned to step 322 and, if possible (N in 322), in step 323 switched to the next queue . Optionally, before step 328, an additional testing step analogous to step 325 could also be processed.

Upon determination that no access commands 19 are listed in the current queue 18xy (N in 324) or that no read access volume quota remainder rCRn is available (N in 326) or that no write access volume quota remainder rCWn is available (N in 328), in a further step 330 checked whether these facts have already been established for all queues 18nm of the current queue group 17x. If this is the case (Y in 330), method step 232 is ended. If this is not yet the case (N in 330), the process returns to step 323 in order to switch to the next queue.

If for the current queue group 17x the access volume quotas Cn are used up without resolving a queue or after resolving one or more queues of this queue group (Y in 322), or also after all queues 18nm of the current queue group 17x have been resolved (Y in 330), method step 232 is terminated, with an identifier of the last processed queue 18z of the current queue group 17x being stored beforehand in step 332.

In a preferred embodiment variant, the in 4 illustrated method step 232 of the first arbitration phase A1 be modified such that in a case when a fragmented access command 19f is detected in phase 310, ie an access command 19 has been fragmented in a previous time frame TR, then no more access command 19 for this queue group 17x is managed if, in the first arbitration phase A1, an access command fragment 19f originating from the previous time frame TR has to be fragmented again due to a lack of sufficient access volume quota Cn, so that in this time frame TR no access volume quota remainder rCn is available for processing further access commands 19 of this access type, or the access volume quota remainder rCn is available completely used up after processing the last fragment.

In general, in this connection a) a fragmentation of access commands 19 can be started in both arbitration phases A1 and A2 by limiting the access volume; b) fragmentation of an access command 19 can only be continued in the first arbitration phase A1 by again limiting the access volume (corresponding to a remaining access quota of zero); and c) a fragmentation of an access command 19 can only be ended in the first arbitration phase A1. At c) he can Access volume quota remainder rCn of the respective access type are used up completely or are available for processing further access commands 19 of this access type in the first arbitration phase A1 and possibly also in the second arbitration phase A2.

Referring to 5 and 6 the second arbitration phase A2 will now be explained in more detail.

In the second arbitration phase A2, in step 510 there is first a wait until all previously permitted access commands 19 from any queues 18nm in any queue groups 17n have been processed. The processing of the access commands 19, 19f permitted in the first arbitration phase A1 may, as far as possible, take place in parallel. Then, in step 512, the remaining time period rtd is calculated until the predetermined time period td of the time frame TR has elapsed, and in step 514 it is checked whether there is still a remaining time period rtd in the time frame TR.

If there is still a remaining time period rtd (Y in 514), in step 516, based on the retrieved identifier of the queue group 17z last processed, the next queue group is determined as the current queue group 17x for processing and then in step 518 this current queue group 17x is processed by a , in particular to allow at most one access command 19 from this queue group 17x. As in 5 illustrated, after a queue group 17x has been dealt with, a return is made to step 510 in order to wait again in the second arbitration phase A2 for the complete processing of permitted access commands and to calculate the remaining time rtd (512) in order to deal with another queue group 17n if necessary (518) .

If there is no remaining time period rtd (N in 514) (after none, one or more queue groups have been dealt with), the second arbitration phase A2 is no longer carried out. In step 520, however, an identifier of the last queue group 17z dealt with is then temporarily stored.

Resolving 518 the current queue group 17x is in 6 illustrated in more detail. First, in step 610, the cached identifier of the last processed queue 18z of the current queue group 17x is retrieved. Then, in step 612, the next queue is accessed as the current queue 18xy of the current queue group 17x for handling based on the retrieved identifier of the last handled queue 18z.

In step 614 it is then checked whether there are any access commands 19 in the current queue 18xy at all. If the current queue 18xy is empty (N in 614), the processing of the current queue 18xy ends immediately. If the current queue 18xy is not empty (Y in 614), it is checked in step 616 whether a read access command 19R is listed in the current queue 18xy. If present (Y in 616), it is checked in step 618 whether there is still a read access volume quota remainder rCRn (i.e. remainder of the allocated read access volume quota CRn in this time frame) for this queue group 17x. If this is the case (Y in 618) (regardless of the value of the read access volume quota remainder), in step 620 the read access command 19R is permitted.

If there is no read access command 19R in the current queue 18xy (N in 616), based on the check result in step 614 that the current queue is not empty (Y in 614), it is assumed that a write access command 19W is in the current queue 18xy is listed. Optionally, an additional testing step analogous to step 616 could also be processed for this. In step 622 it is therefore checked whether there is still a write access volume quota remainder rCWn (i.e. remainder of the allocated write access volume quota CWn in this time frame) for this queue group 17x. If this is the case (Y in 622) (regardless of the value of the read access volume quota remainder), in step 624 the write access command 19W is allowed.

As in 6 shown, in the second arbitration phase A2 in this context (in contrast to the first arbitration phase A1) at most one access command 19 of a queue 18 nm is permitted. Relating to 5 it can also be seen that after an access command 19 has been admitted and processed or all queues 18n of the queue group 17x have been searched for executable access commands 19 without result, the next queue group 17x+1 is searched for executable access commands 19. In other words, in the second arbitration phase A2 (likewise in contrast to the first arbitration phase A1), in each round of administration of the queue groups 17n only a maximum of one access command 19 per queue group 17n is permitted (the decisive method steps are 616, 626, 630). However, the queue groups 17n are cyclically changed according to access commands 19 searched (ie also in several management rounds via the queue groups 17n) until the end of the time frame TR is reached or in all queue groups 17n there are no longer any access volume quota residues rCn.

After the read access command 19R has been allowed in step 620 or the write access command 19W in step 624, it is first checked in step 626 whether processing of the allowed access command 19 has started. If this is the case (Y in 626), in step 628 an identifier of the last processed queue 18z of the current queue group 17x is stored.

After determining that no access commands 19 are listed in the current queue 18xy (N in 614) or that no read access volume quota remainder rCRn is available (N in 618) or that no write access volume quota rCWn is available (N in 622) or that the processing of the permitted access command 19 does not start (N in 626), it is checked in a further step 630 whether all queues 18nm of the current queue group 17x have already been dealt with. If so (Y in 630), the process proceeds directly to step 628. If all queues have not yet been dealt with (N in 630), the process returns to step 612 to move to the next queue 18xy of the current queue group 17x switch. In this context, it should be noted that the processing of the permitted access commands 19 should take place sequentially in the second arbitration phase A2 (while this can possibly take place in parallel in the first arbitration phase A1).

Referring to 7 and 8th a preferred embodiment of an admission process 700 of a read access command 19R or a write access command 19L with a possible fragmentation of the access command 19, if necessary, will now be explained in more detail. The admission process 700 refers to the in 4 indicated steps 313, 316, 327 and 329 in the first arbitration phase A1 and in 6 indicated steps 620 and 624 in the second arbitration phase A2. In other words, access commands 19f that were fragmented in the previous time frame TR can be fragmented again in the first arbitration phase A1 if the remainder of the fragmentation cannot be completely permitted. Such a second fragmentation of an access command takes place only in the first arbitration phase A1, whereas a first fragmentation of an access command can take place both in the first and in the second arbitration phase A1, A2.

In the first step 710, the access command 19 to be permitted may be limited based on the access volume quota Cn. As in 8th illustrated, it is first checked in step 710a whether the access volume ZV of the access command 19 is greater than the access volume quota remainder rCn of the corresponding allocated access volume quota Cn. If this is not the case (N in 710a), the access command can be processed in full in relation to quotas and therefore does not have to be limited. If, on the other hand, the access volume ZV of the access command 19 to be permitted exceeds the access volume quota remainder rCn (Y in 710a), then the access volume ZV is limited in step 710b to the access volume quota remainder rCn that is still present. This quota-related limit check is performed equally for read access volume quotas CRn and write access volume quotas CWn.

After this quota-related limitation, which may have to be carried out in step 710, it is checked in step 712 whether a duration-related limitation has to be taken into account at all, which is only the case in the second arbitration phase A2. The duration-related limitation in step 714 can be necessary without prior quota-related limitation, but can also be applicable to an access command that is already quota-related.

In the second arbitration phase A2 (Y in 712), in step 714 the access command 19 to be permitted may be limited (possibly a further limitation of the access command that is already limited in relation to quotas) based on the predetermined time period td of the current time frame TR. Here, in particular, very precise consideration is given to how long it takes to process an access volume ZV of the respective access command 19 . It should be noted, for example, that different types of access can require different access times (e.g. write accesses with the same volume usually take significantly more time than read accesses). It should also be noted, for example, that access times for volumes generally cannot be calculated linearly, since, for example, an access time for 128 kB is generally significantly less than eight times the access time for 16 kB. In order to carry out this limiting step more simply and reliably, a table can preferably be accessed in which the connections between access times and access volumes are listed. If the access time for processing the access command 19 is shorter than the available remaining time rtd of the current time frame, the access command can be processed in full and therefore does not have to be limited in terms of time. If, on the other hand, the access time for processing the access command 19 is longer than the available remaining time rtd of the current time frame, then the access volume ZV is limited to a level for whose access time the remaining time rtd of the time frame TR is sufficient. This time-based limitation check is also carried out equally for read access volume quotas CRn and write access volume quotas CWn.

Such a limitation of the access command to be processed in a time frame leads to a corresponding fragmentation of the access command over at least two time frames.

Next, in step 716, it is checked whether the access volume ZV has been limited to zero, i.e. whether limited processing of the access command 19 is not even possible. If this is the case (Y in 716), i.e. if no (not even a limited) processing of the access command 19 is possible within the remaining period rtd of the time frame TR, then the admission process 700 is completed directly, without processing or at least a limited one Allow access command 19 to be processed.

Otherwise (N in 716), the approval process 700 continues. In the next step 718 it is checked whether in steps 710 and/or 714 the access command 19 was limited. If this is the case (Y in 718), in the next step 720 it is checked whether the now limited access command is an access command that was already fragmented in the previous time frame. This means that access commands can be fragmented multiple times over a number of time frames TR. Accordingly, in step 722a the access command 19 that is currently fragmented for the first time is stored in the corresponding queue group 17x, or in step 722b the storage of the previously fragmented access command 19 in the queue group 17x is updated, so that the now fragmented access command is recognized as such in the next time frame TR and is preferred as first access command can be processed in the first arbitration phase A1.

Then, in step 724, the access volume value consumed by the access command 19 is adjusted to the limitation that has been set. And after that, in step 726, the access volume quota remainders rCn are updated according to the limitation. If no limitation has taken place (N in 718), the access volume quota remainders rCn are also adjusted in step 726, but of course based on the full access volume ZV of the access command. And then, in the last step 728, the processing of the permitted (limited or non-limited) access command is started.

The maximum available bandwidth is distributed between the several applications and used very efficiently with this described method of managing access to the data memory. This is achieved in particular by the two special different arbitration phases and is preferably executed even more efficiently by the possible fragmentation of the access commands.

The above referring to 1 until 8th The embodiment described by way of example is intended to provide a better understanding of the invention as defined in the appended claims, without limiting the scope of the claims. The invention defined in the claims can also be implemented using additional or modified features and also by omitting individual features described.

Reference List

10

computer system

12

processor

14

software system

15n

applications

16

host software

17n

queue groups

17x

current queue group

17z

last processed queue group

18nm

Queues in queue groups

18xy

current queue of the current queue group

18z

last processed queue

19

access command

19f

fragmented access command

19R / 19W

Read access command / write access command

19fR / 19fW

fragmented read access command / write access command

20

data storage

22

data access element

24

access controller

25

program memory

200

Access Management Procedures

A1

first arbitration phase

A2

second arbitration phase

cn

Access volume quota for 17n in TR

rCn

Access Volume Quota Remainder for 17n in TR

CRn

Read access volume quota for 17n in TR

CWn

Write access volume quota for 17n in TR

TR

time frame

td

predetermined length of time of the time frame

rd

Remaining time in the time frame

ZV

access volume

Claims (15)

Method (200) for managing access from multiple applications (15n) to a data memory (20) via time frames (TR) with predetermined time periods (td), the multiple applications (15n) each having at least one queue group (17n) during their execution generate in each case at least one queue (18 nm) for listing access commands (19) to the data memory (20), the method having: acquiring (210) individual access volume quotas (Cn) each allocated to one of the plurality of queue groups (17n), wherein the sum of the access volume quotas can be fully consumed within a predetermined period of time (td) of a pending time frame (TR); starting (220) the time frame (TR) with the predetermined time period (td); Carrying out (230) a first arbitration phase (A1), in which in a single management round in succession in the queue groups (17n) of all applications (15n) in each case in all queues (18nm) independently of a remaining period of the time frame (TR) at least one of the listed Access commands (19) are permitted if the respective queue group (17n) allocated access volume quota (Cn) has not yet been used up; and Carrying out (240) a second arbitration phase (A2), in which in several management rounds in succession in the queue groups (17n) of all applications (15n) at most one of the listed access commands (19) is permitted if the respective queue group (17n) allocated access volume quota (Cn) has not yet been used up, the second arbitration phase (A2) being terminated when the predetermined period of time (td) of the time frame has elapsed. Method (200) according to claim 1 , in which the queues (18nm) of the queue groups (17n) are each configured to list access commands of a first access type and/or access commands of a second access type; and the access volume quotas (Cn) allocated to the plurality of queue groups (17n) each include a first access volume quota (CRn) for the first access type and/or a second access volume quota (CWn) for the second access type. procedure after claim 2 , in which in the first arbitration phase (A1) in the one management round in the queues (18nm) one of the listed access commands of the first access type is allowed as long as the respective queue group (17n) allocated access volume quota (CRn) for the first access type has not yet has been used up, and one of the listed access commands of the second access type is permitted as long as the access volume quota (CWn) allocated to the respective queue group (17n) for the second access type has not yet been used up; and in the second arbitration phase (A2) in the several administration rounds only one of the listed access commands of the first access type or one of the listed access commands of the second access type is permitted in each of the queue groups (17n). Method (200) according to one of the preceding claims, in which, if a residual value of the access volume quota (Cn) allocated to the respective queue group (17n) is not sufficient for the complete execution of an access command (19), the access command (19) is fragmented in such a way that allow processing of only part of the access volume of the access command corresponding to the remainder of the allocated access volume quota (Cn). Method (200) according to one of the preceding claims, in which, if in the second arbitration phase (A2) a residual time period of the time frame (TR) is not sufficient for the complete execution of an access command (19), the access command (19) is fragmented in such a way that allow only part of the access volume of the access command to be processed, des sen access time corresponds to the remaining time of the time frame. Method (200) according to claim 4 or 5 , in which an identifier of the fragmented access command (19f) is temporarily stored and a next processing of the queue groups (17n) begins in a next time frame (TR) with the admission management of the access command (19f) fragmented in the previous time frame (TR). Method (200) according to any one of Claims 4 until 6 , if dependent on claim 3 , in which no further access command (19) of a queue (18nm) is managed as long as an access command taken from this queue in fragments cannot be processed completely in the first arbitration phase (A1). Method (200) according to one of the preceding claims, in which the time frame (TR) is terminated before the predetermined time period (td) has elapsed if the access volume quotas (Cn) of all queue groups (17n) have been completely used up. Method (200) according to one of the preceding claims, in which an identifier of the last processed queue (18nm) of a queue group (17n) is buffered in each case and a next management phase of the same queue group (17n) begins with that queue after the last processed queue; and/or an identifier of the queue group (17n) last dealt with is buffered in each case and a next management phase begins with that queue group after the queue group last dealt with. Method (200) according to one of the preceding claims, in which, to determine whether a remaining period of time in the time frame (TR) is sufficient for processing an access command (19), an access time claimed by the access command (19) is determined based on a table in which the connection between access times and access volumes is listed. The method (200) of any preceding claim, wherein obtaining (210) the access volume quotas (Cn) comprises retrieving the access volume quotas (Cn) from host software (16) associated with the plurality of queue groups (17n) of the plurality of applications ( 15n) allocates the access volume quotas (Cn), the sum of which can be fully utilized within the predetermined period of time (td). Method (200) according to one of the preceding claims, in which the access management in the second arbitration phase (A2) only begins after the completion of the processing of all access commands (19) permitted for processing from the first arbitration phase (A1). A computer system (10) comprising: at least one processor (12); a data store (22); several applications (15n), each of which can be executed on the at least one processor (12) and during their execution at least one queue group (17n) each with at least one queue (18nm) for listing access commands (19) to the data memory (20) generate; and an access controller (24) which is configured to manage the accesses of the plurality of applications (15n) to the data store (20) according to the method according to any one of Claims 1 until 12 . Computer system (10) according to Claim 13 , further having host software (16) which can be executed on the at least one processor (12) and which, when it is executed, allocates access volume quotas (Cn) to the plurality of queue groups (17n) of the plurality of applications (15n), the sum of which is within a predetermined Duration (td) of an upcoming time frame (TR) can be fully utilized. Computer-readable storage medium (25) containing instructions that can be executed by an access controller (24) of a computer system (10) and when they are executed the access controller (24) the method (200) for managing access from multiple applications (15n) to one Data memory (20) of the computer system (25) according to one of Claims 1 until 12 run.

Images