Detection of Sequential Data Stream

This invention relates to the field of data storage and management. More particularly, the invention relates to a method for detecting whether a particular data set having been requested from long term storage by a host computer is likely to be part of a sequence of such sets, and for taking appropriate action in response to this determination.

Modern data processing systems typically comprise a host computer, consisting of an arithmetic and logic unit and a main memory unit for containment of data and instructions presenting being processed, and long-term storage means for storage of data and processing instructions at other times. In systems using the IBM Corporation's equipment, the long-term storage means is connected to the host computer by means of a "channel." When the host desires a particular data set or record, it issues a command over the channel to the long-term storage means, which then locates and reads the data from whatever medium it is stored upon, e.g., magnetic disks or tape memory media, over the channel into the main memory of the host. The substantial length of time required to retrieve data from long term storage limits the throughput or usage of the host computer. In particular, location of the beginning of the data set, e.g., physical juxtaposition of the location of the beginning of a record stored on disk to the read/write head, is time consuming. The actual reading of the data proceeds comparatively quickly. To minimize this loss of use of the host computer, the host will typically issue a series of requests for data and then perform other tasks while the data is being retrieved from long term disk or tape memory. However, even when this "queueing" is performed there is substantial host computer computation time lost due to the time required for accessing data and software overhead associated with the queueing process. This has remained an unsolved problem in the art and it is an object of the present invention to improve host computer throughput by reducing queueing times.

It has been proposed in the prior art that such queueing times be reduced by "staging" data physically stored surrounding all data which is the object of a SEEK command issued by a host, from a disk memory into a solid-state memory of much faster access speed. The solid-state memory is located external to the host, outboard of the channel from the host. Thus, when the host issues subsequent READ commands, the data sought may already be contained in the high speed solid-state memory and can be supplied to the host more or less instantaneously. However, if all data sets surrounding records accessed by the host are read into a solid-state memory external to the host as described above, the problem of queueing is not entirely eliminated, as then the channel and director usage time consumed while data is read into cache memory is added to the actual latency time required for the data set to be located on the disk and juxtaposed to the head.

Moreover, it will be appreciated that there are generally two ways in which data is accessed by a host computer. All the data in a given data set may be called for by the host at a given time, or the host may initiate a separate call for each portion of the data set as required. In the first case, addition of the cache memory to the system adds no performance improvement, as but a single latency time is required to satisfy each input/output request. In the second case, wherein each individual host instruction is part of a sequence of instructions typically directed to access successive portions of a physical record such as a tape or disk drive, latency time is consumed in responding to each portion of the data set. In this situation, the total latency time can be reduced to that of a single access operation if successive portions of the data set are read into a high speed solid-state cache. Subsequent requests for other portions of the data set can then be satisfied directly from solid-state memory without involving second and successive physical access operations. That is, if the data is cached in anticipation of a subsequent SEEK command, it will be available immediately. Accordingly, it is desirable that means be provided for determining which data requests made by a host computer are likely to be part of a sequence of such requests.

It would not, of course, be impossible for the host computer to issue a single indicating whether or not a particular data set called for is part of a sequence of such sets, and some systems now being announced will have this feature. This would, of course, simplify the decision as to whether or not to "stage" the subsequent record from the long-term data storage means into a cache memory. However, many existing computing systems of commercial importance (such as most of the IBM Corporation's line of computers) do not provide such a signal. Nor is it desirable to modify these computers, in particular their operating systems, in order to provide such a signal as such modifications are difficult to implement correctly and are not popular with computer users.

Accordingly, it is desirable to render the caching of data function more efficient by using improved means and methods to determine whether a particular data request made by a host computer is part of a sequence of requests directed to the same data set (in which event the subsequent portion of the data set would be cached) while data which is not amenable to efficient caching is processed in the same manner as in the prior art.

It is a further object of the invention to provide a system in which sequential portions of a data set can be cached so as to improve thoughput of a host computer system, without requiring modification to the host.

Yet another object of the invention is to provide a means and method for detecting whether or not a data record sought is part of a sequence of such records, wherein the means and method operates using information contained within the "channel program" processed by the storage detector, whereby implementation of the method of the invention is rendeted simple and relatively inexpensive.

A further object of the invention is to provide a method whereby an area in the cache assigned to a particular data set can be deallocated therefrom automatically so as to free storage space for reuse.

The present invention satisfies the needs of the art and objects of the invention as noted above by examining each data record that is read from long-term data storage means to a host computer. Control signals contained in the (conventional) channel program are used to provide negative indicators, i.e., signs that the record sought is not one of a sequence of records stored together on long-term storage means. If none of these indicators is found during the retrieval of a given record from long-term storage to a host computer, succeeding records are then cached. Upon the receipt of each successive "read" request from the host computer, the request is compared with a directory which keeps track of which records have been cached so that the host's request will be satisfied from the cache if possible. In a particularly preferred embodiment, if a first record on a given disk track is read from the disk storage device into the host and no indicator that the first record is not part of a sequence of records on the given track is detected, then the data on the given track is cached. If the data on the rest of this track is then called for by the host computer, the track next succeeding the given track is then cached, again without host intervention, and so on. A track may also be "prestaged" if, for example, there are not enough unread bytes in the cache to satisfy a subsequent data request equal in size to that most recently satisfied.

In a further preferred embodiment, at the time of installation of the computer system or at a later date, certain disk packs, i.e. physical assemblies of magnetic storage disks, may be expected to contain substantially random data sets, while others may be expected to store mainly sequential data sets. Upon a host's request for data stored on one of these disk packs, differing control sequences may be undergone to determine whether caching of further data would be desirable. For example, if the access request is to a disk pack considered to be substantially random, a comparatively lesser amount of data may be staged to the cache which is useful if the request turns out to be one of a sequence. If the request is to a disk pack considered to contain mainly sequential records, a substantially larger amount of data can be cached in response to the determination that caching would desirable.

The invention will be better understood if reference is made to the accompanying drawings, in which:

FIG. 1 represents a simplified view of the system within which the invention forms a part;

FIGS. 2-5, 8 and 9 represent flowcharts of the decision-making process according to the invention; and

FIGS. 6 and 7 show graphically the performance improvements made by the invention.

The method of detection of a sequential data stream of the invention was designed in particular to operate within the cachhe memory buffered subsystem which is the subject of copending application Ser. No. 325,346 filed Nov. 27, 1981. The chief objective of the cache memory buffered subsystem which is the subject of the copending application referred to above and of which the present invention forms a part is to save on latency time required to complete a SEEK instruction given by the host computer by reading data into a faster access, typically solid-state memory, from a slower access, long-term magnetic storage device, in advance of the host actually calling for the data. In this way, if the cached record is then called for, a physical delay need not be undergone. Clearly, it is desirable that only those records which, in fact, are portions of sequentially accessed sequences of records be staged to the solid-state memory or "cached." Otherwise, caching of data will have beneficial effects only when substantially all the records sought for are, in fact, sequential, e.g., on the order of 90%. This is not the case in typical commercial systems. Accordingly, it is required that some means be provided to detect which records are portions of sequences of records and to cache only those. The present invention provides means for making the determination.

As discussed above, in the preferred embodiment of the invention, some data may be staged even when it is not clearly determined that a given request is part of a sequence of requests; simulation testing shown that on a statistical basis, this will still provide a performance improvement.

The invention is designed to operate in a cache buffered memory subsystem operating as an add-on to a pre-existing memory system which is plug-compatible with a host computer; in a preferred embodiment, with an IBM 370 Series machine operating under a typical IBM operating system. The board layout of the system of the copending application referred to above is shown in FIG. 1. There, the host computer 10 passes commands and data over and receives data from a channel interface indicated generally at 12. In the event that the data to be accessed is stored on a disk drive such as at 14, access commands are passed to a storage director 16 which converts it into a form suitable for controlling a control module 18 which actually controls the movement of the disk drives 14, e.g., the radial movement of read/write heads with respect to the disks and the like. The system as thus far described is conventional. What is added by the invention of the copending application referred to above is a cache memory subsystem indicated generally at 20 comprising a solid-state cache memory 22 and a cache manager 24 for determining which data is to be staged into the cache memory 22 from the disks 14. It is accordingly the cache manager 24 which carries out the method of the present invention, which will now be described in detail.

When the host computer 10 seeks to obtain data from a disk storage system 14, it issues a well defined SEEK command over the channel 12, which is acted upon by the storage director, in processing the so-called "channel program", to cause the control module 18 to access the data sought for. The data is then passed back through the control module 18 and the storage director 16 and over the channel interface 12 to the host 10. The actual reading and writing of data is performed under the channel program, including the SEEK command, defined by IBM, as is the channel interface itself, in a document entitled "I/O Interface Channel to Control Unit", IBM Publication GA22-6974-4. As will be appreciated by those skilled in the art, this well defined protocol permits the channel program to contain additional instructions for further data reading and writing. For example, if a given data set is stored at several different locations on disk, for example, on several different "cylinders", these being areas of data storage spaced at varying distances from the central axis of the disk stack, a CYLINDER SEEK command may be encountered during the channel program. The control module and storage director then cause the head to access the new cylinder sought for and to continue the reading operation. Similar instructions include HEAD SEEK commands. Thus, typically in the prior art the storage director will follow such additional "embedded" commands encountered in the channel program. Inasmuch as the system of the present invention is very desirably "software transparent" and "plug compatible" to the host, so that no modification of the host is required, it continues to obey these instructions. However, the cache manager comprises means for examining information received from the director in performance of the channel program, for determining whether additional data should be staged from the disk to the cache after completion of the actual operation ordered by the host. Specifically, the cache manager 24 examines the channel commands as they pass through the storage director for commands such as those above, e.g. CYLINDER SEEK, HEAD SEEK, and the like. If it receives no indications that the data is not of a sequential nature--these being defined more fully below--it then continues with the reading of all the data stored on a given track (i.e., an area of data storage on a single surface of a disk one full revolution long) into the cache. If the host then calls for the remainder of the track, the cache manager detects this and causes the next track to be similar staged to the cache without host intervention, and so on. In this connection, note that the "next track" may not be physically adjacent to the last on the same disk surface. Depending on the organization of the disk system, the "next track" may be on an entirely different disk surface, though usually one located above or below the last in a stack of disks, so as to save on both seek and latency delay times.

Accordingly, the cache manager examines the channel commands throughout the read channel program, while the actual decision as to whether the following track should be staged is made at completion of the request; that is, at the conclusion of the conventional "channel program." The present invention therefore relates to additional decision-making, carried out independent of the host, by the cache manager 24. Specifically, the storage director 16 reports to the cache manager 24 certain information or "statistics" which enable the cache manager 24 to make decisions concerning the desirability of staging successive tracks. In a presently preferred embodiment, these statistics may include:

1. Was the first record on the track among the records processed?

2. Was the last record on the track among the records processed?

3. Was the "End of File" record among the records processed?

4. Was the channel program processed against the cache or direct to the disk storage device?

5. Were cahce or disk read errors encountered?

6. Was an embedded head or cylinder seek command encountered in the channel program?

7. Did a write command occur during the channel program?

8. How many records were processed in the track during the channel program? and

9. How long were the records processed?

As discussed above there are two basic types of decisions relating to staging to be made following execution of the channel program. First, if the channel program was processed directly from the storage director to the disk drive, i.e., a non-cached record was read, the question as to whether the remainder of the track should be staged into the cache arises. The second question arises when the channel program was processed to the cache, that is, the data sought for by the channel program was indeed found in the cache. In this case, the question is whether it is desirable to "prestage" the entire next disk track into the cache. Of the above statistics, the following disquality data from being cached, both as to stage and prestage: errors, write commands, or "End of File" records. These are all counterindicators of the desirability of caching, as they indicate that the next record on the disk track is not necessarily part of the same data set as that read, or that some part of the system is malfunctioning.

Embedded seek commands are also considered as part of the statistics, but they are treated somewhat differently depending on the embodiment of the invention. As discussed briefly above, in a presently preferred embodiment of the invention, certain individual disk drives may be determined to contain largely sequential data sets or largely random data sets. In either case, caching may be of some improvement. For example, in the random case, the host may not always call for all the data in a given data set at one time, so that it would still be desirable to do limited caching, in the case of a READ command directed to such a disk pack. Encountering a CYLINDER SEEK command during the processing of a channel program would counterindicate the desirability of caching because presumably even a sequential data set would not often extend across plural cylinders on a disk pack. However, a HEAD SEEK, indicating that the next record sought for by the channel program is in the same cylinder as that previously read, is not considered a counterindication of the desirability of caching even in the random data set.

HEAD SEEK and CYLINDER SEEK commands are treated differently when the disk pack is considered to be likely to contain sequential data. In the case of sequential data, both HEAD SEEKs and CYLINDER SEEKs counterindicate caching because there is no insurance that subsequent data sets may not be processed to the same cylinder. Essentially, in the case of a random disk pack, caching is performed without regard to sequentiality, so HEAD SEEKs do not preclude caching. The amount of data cached is kept deliberately small so as to minimize the impact of unnecessary caching which does not lead to a direct benefit.

FIG. 2 shows the decision-making process to be undergone at the completion of each read operation, i.e., upon completion of the channel program, at 28, with respect to a disk pack considered to contain mainly sequential data sets. The examination of the statistics kept by the director to determine whether any of the statistics disqualify the record from staging is performed first, as indicated at block 30 where the existence of any of the "no stage" indicators, items 3 and 5-7 of the statistics listed above, is determined. If there are such indicators the data is not staged as noted at 32. If there are no such indicators, the next question is whether the previous access was directly from the disk to the host or whether the operation was from the cache to the host, as indicated at decision block 34. If the operation was from disk to host (NO answer to block 34) the data is looked at to see whether the first record on the track was read, and that not more than 38,138 bytes (the maximum number of bytes stored on two tracks of a particular mode of disk drive) were read as shown at 36. If so, then at least the remainder of the track is staged, at 38. Under certain circumstances, as discussed below in connection with FIG. 3b, several more tracks are staged.

Returning to consideration of block 34, if the previous operation was from the cache to the host, indicating that a sequence of records has been correctly detected and cached, then the question is whether prestage of the next track is desirable. At decision block 40 the question is whether prestage is allowed for records stored on a particular disk device. This allows the system designer to prevent the data stored on certain disk devices from being cached, i.e., allows him to save certain disk for data he knows will be random. For example, so-called "fixed head disks" which typically have less latency time than movable head disk drives would ordinarily not be used for sequential data sets. If prestage is not allowed, then the decision is made simply not to stage. If prestage is permitted, the cache manager 24 (FIG. 1) then determines if prestage is desirable. For example, the decision might consider whether there were enough bytes remaining in read in the cache to satisfy a subsequent data request of the same size, as noted at 42. Similarly, the manager might look at the length of the record requested in a given request, and stage more than the remainder of the track, or more than a single following track, if it appears that one or more subsequent similar requests will not otherwise be satisfied by the data staged to the cache. If both conditions are satisfied, a prestage subroutine described in connection with FIG. 3a is entered, at 44.

FIGS. 3a and 3b show how the number of tracks to be staged or pre-staged is determined. In both cases, the number of tracks to be staged or pre-staged is proportional to the number of bytes read in performance of the channel program. The assumption is made that the longer the request, the more likely it is that subsequent requests will also be long. Thus, in FIG. 3a if less than 6000 bytes are read, one track is prestaged, if between 6000 and 12,000 bytes were read two tracks are prestaged, while if the total number of bytes is between 12,000 and 38,138 three tracks are staged. (As noted above, the number "38,138" will be recognized by those skilled in the art as being the maximum number of bytes stored on two disk tracks on a popular family of disk drives.) Accordingly, if a single READ request totalling two tacks was performed, three tracks are prestaged to the cache, thus insuring that similar subsequent requests can be satisfied without incurring a latency time penalty. If the total number of bytes read in a particular operation is more than two tracks, i.e., more than 38,138 bytes, prestaging to the cache is not performed. This is so simply because satisfaction of a caching request of this length will consume considerable director and cache manager operation time which might be better expended elsewhere.

FIG. 3b shows the similar calculations for staging. One track is staged if the total bytes read in the execution of the channel program was under 4500; two if it was between 4500 and 9000, and three if it was between 9000 and 38,138.

It will be appreciated that it is important to keep track of the use of each of the tracks of data stored in the cache so that when data need no longer be stored in the cache the "frame"--i.e., the area of cache memory--assigned to that data can be "deallocated" and returned to the pool of empty frames. Deallocation is performed in order to free storage space for storage of data (discussed in connection with FIG. 4) and to ensure data integrity (FIG. 5). FIG. 4 shows a flowchart of operations undergone in deallocation operations. It begins at 50 with the question, "EOF record read?" If the EOF (end of file) record, i.e., the last record in a given data file, has been read, clearly it will not be desirable to do any further staging, nor will there be any anticipated further use for the data. Accordingly, the frame assigned to that track is deallocated at 52. Likewise, if an indication is found at 58 that the last record of a track was read, there is no need to preserve data stored in the preceding frame, i.e., the frame containing the data from the current track address minus 1, so that it is deallocated as at 60.

Performance of the deallocation subroutine shown in FIG. 4 is followed by that shown in FIG. 5, which as noted above, controls deallocation for data integrity purposes. Thus, for example, if the host writes to a cached track at 70, clearly the cache area in which the data corresponding to the write operation is stored should be deallocated as no longer containing valid data. Decision blocks 72 through 80 indicate how much of the cache should be deallocated upon indication of a write operation. For example, if at 72 an embedded seek was encountered, the cylinder should be deallocated at 74. If a multitrack operation, i.e., one in which writing was performed to a sequence of tracks, occurred as at 76, that range of tracks is deallocated at 78. Otherwise, simply the single track to which the write occurred is deallocated at 80. If the channel program ran only a read operation, i.e., the decision in block 70 has a negative result, then one simply looks to see whether a cache read error was detected during the reading at 82. If so, the frame is permanently deallocated at 84, on the assumption that a defective memory element is contained therein or the like.

FIG. 8 shows the flowchart of the decision-making process undergoen in the presently preferred embodiment of the invention in which data stored on a disk pack anticipated to contain mainly random, but possibly some sequential data sets, is considered for caching. A fixed amount of data is cached in response to all read requests; the amount of data cached is determined by weighing savings in latency time realized thereby versus the director time consumed by staging data. At present the amount of data anticipated to be staged is one full track. Even if the data request turns out to have been actually random, this practice does not unduly burden the system with uselessly cached data, and may give a performance improvement. Computer modelling assuming a track size of 19k bytes and a total cache space of 12m bytes indicates that an improvement will exist. The module begins at 86 by considering whether any of the no stage indicators are on; if so, no staging is done. If the total bytes read is greater than 6,000, no staging is undergone, because data blocks of this size consume substantial director time in caching. The track is staged at 90. If record one of the track was also read, at 92, a "possibly sequential" flag is set at 94. If this flag is set, prestage determination is undergone next. This is shown in FIG. 9. An attempt at 96 is made to deallocate the track having the current track address minus two (CTA-2 ), i.e., the track two tracks preceding that presently being considered, in the conventional sequence of reading. If this is possible, at 98, which indicates that 3 tracks in a row have been read, then the following track address, track CTA+1, is prestaged at 100, again assuming that the sequence is likely to continue. Deallocation is performed with respect to data cached from a random disk pack in substantially similar fashion as discussed above in connection with sequential disk packs. Deallocation of the current track address-2 is performed at 96, as noted above, and the data integrity deallocation routine described in connection with FIG. 5 is undergone as well, in particular with respect to cache read errors.

It should be noted as well that the deallocation is performed with respect to random disk packs in accordance with allocation of frames to data to be staged, in accordance with co-pending application, Ser. No. 325,351 entitled "Adaptive Domain Partitioning of Cache Memory Space", now U.S. Pat. No. 4,430,712, and a co-pending continuation-in-part application Ser. No. 441,791, filed Nov. 15, 1982 and now U.S. Pat. No. 4,503,501 having the same title. In accordance with that invention, random data cached according to the methods discussed in connection with FIGS. 8 and 9 is eventually deallocated simply on the basis of not having been used.

The flowcharts of FIGS. 2-5 and 8-9 will be appreciated by those skilled in the art as ones designed to operate as subprograms within a cache manager control program. In particular, the methods of the invention as described above are designed to operate within the Model 8890 disk controller of Storage Technology Corporation, the assignee of the present invention. Accordingly, the decision-making sequences shown in FIGS. 2-5 are configured as subprograms undergone every time a channel program is executed. There are numerous other ways in which it would be possible to implement the method of the invention. For example, staging might be considered only where the record most recently read began a track and prestaging could only be considered at the end of a track. Those skilled in the art will recognize numerous other expedients are possible.

In a presently preferred embodiment, the decision-making capacity of the invention is embodied in microcode permanently written on a floppy disk physically carried within the storage director unit and copied to solid-state random-access memory each time the director is turned on. This is operatively connected to a preexisting storage director processor for performing the cache manager function.

FIGS. 6 and 7 show graphically the performance improvements provided by the caching subsystem of the invention. These figures represent data generated by computer simulation of system performance. No comparable actual data is available as of the date of filing of the instant application though testing with respect to individual jobs indicates that a substantial performance advantage is realized. These results related to a preferred embodiment in which data is cached in units of disk tracks, which for modelling purposes, are conceptually defined to contain on the order of 4 records.

FIG. 6 shows average input/output command response time versus the block size of the data being accessed. "Input/output (I/O) time" includes the latency time required for a particular sector of a given disk to be juxtaposed to the read/write head and the time required for the read/write head to move radially with respect to the disk (in those disk drives where movable heads are used), in addition to "RPS miss time", time lost due to components of the I/O path not being available at the time the data is available. Seek and latency delays are independent of the block size. However, since the longer block requires more time to be written or read, input/output transfer time and RPS miss time both increase as the block size increases. Curve A of FIG. 6 shows this, graphically representing the average response time of a prior art system, i.e., one in which no cache is present. There the longer the block the longer the average response time. The minimum time reached is non-zero, as latency and access time is required even for a block of extremely short size.

Curve B of FIG. 6 shows the improvement made by caching a larger portion of data than that requested upon receipt of all input/output requests; that is, no distinction is drawn between those requests deemed likely to be sequential and those which are random. The input/output time with respect to smaller block sizes is improved, as the latency time is significantly reduced. However, the fact that data is cached even when non-sequential, i.e., in response to random requests, thus consuming additional time due to director unavailability, means that the average input/output time with respect to large blocks is greater than in a non-cached subsystem. Hence, curve B shows that the average input/output time is higher than for the non-cached device of Curve A as the block size increases.

Curve C of FIG. 6 shows the improvement made by the present invention where a distinction is effectively drawn between data which is deemed likely to be sequential and that which is not. Since randomly accessed data, i.e., data not identified as being sequential, is not cached, random data has the same input/output access time as in the prior art as represented by curve A. The improvement made by caching sequential data shows up in that curve C is disposed beneath curve A until the block lengths get quite long, at which point, as noted above, latency is of less significance than the actual read/write time required.

FIG. 7 shows average input/output time as a function of the percentage of the data called for by the host which is sequential. All the curves assume a relatively busy system. Curve A again shows the performance of a non-cached disk system. No distinction is drawn in such a system between sequential and randomly accessed data, and hence curve A is substantially flat. Curve B represents a cache system in which all data requests are cached. It shows a marked decrease of input/output time as the percentage of sequential data increases, because of less random data is cached unnecessarily, the percentage of sequential data beneficially cached increases, so that performance is improved as reflected in the lowering of the input/output time. However, the percentage of data which is sequential must be quite high before a performance increase is realized. Curve C again shows the input/output time of a cache memory subsystem according to the invention in which only data deemed likely to be sequential is cached. The overall performance improvement shown is attributable to the fact that the cache system is only used in the case of sequential data; director time is not consumed uselessly by caching random data.

As an example of the methods of implementation of the invention, attached hereto as Appendix A is a "pseudocode" rendition of the method of the invention. This pseudocode includes both actual microcode instructions and documentary notes so as to enable one other than the original programmer to understand the program. A document entitled "Chinook Prolog and Pseudocode Conventions" is attached as Appendix A-1 and explains the pseudocode language. Appendix A describes a microcode module entitled "End-Channel Program Processing." It begins with a general description of the program function, and with operational descriptions including the bits which are required from the storage director for its operation, definition of the various subroutines and of the various registers required. The bulk of the program begins at line 01820 (page 4) and comprises actual microcode instructions, for example at lines 02170 through 02370, preceded by English language explanatory matter describing the subject matter of the decisions made by the subsequent microcode, e.g. as at lines 1980 through 2150. The microcode shown performs the overall management function, and calls various subportions as necessary. For example, the deallocation for data integrity subroutine which is described above in connection with FIG. 5 is called at lines 2440-2470 of Appendix A. The actual subroutine is attached as Appendix B. Of this, for example, the steps taken in response to the decision represented by the box 80 in FIG. 5, "Cache Read Error" are shown in English language form at lines 1500 through 1570 and in microcode at lines 1590 through 1730 of Appendix B.

Those skilled in the art given the thus incorporated documentation of the End-Channel Program would have no difficulty in practicing the process of the invention.

Other copending applications which may be of relevance in understanding the present invention are incorporated herein by reference, including copending Ser. No. 325,351 filed Nov. 27, 1981, of Coulson et al entitled "Adaptive Domain Partitioning of Cache Memory Space", now U.S. Pat. No. 4,430,712, which discusses the organization of the memory space comprised in the solid-state cache memory 22 and a continuation-in-part of that application Ser. No. 06/441,791, filed 11/15/82, now U.S. Pat. No. 4,503,501. Error management and recovery functions performed by the cache manager are discussed in copending Ser. No. 325,287 filed Nov. 27, 1981, of Blickenstaff et al, entitled "Data Path Reconfiguration for Error Control", incorporated herein by reference. Also relevant is copending application Ser. No. 325,288 filed Nov. 27, 1981 of Moreno et al entitled "Control of Cache Buffer For Memory Subsystem" and a continuation-in-part of that application Ser. No. 06/441,818, filed 11/15/82, incorporated by reference, which discusses how precise correspondence between storage locations on the disk devices and the cache is maintained.

Those skilled in the art will recognize that there has been described a method for determining the sequential nature of a disk record which satisfies the needs of the art and objects of the invention discussed above. Moreover, it will be recognized that there are numerous ways in which the method of the invention can be embodied and improved upon and that therefore the above disclosure should not be considered as a limitation on the invention but merely as exemplary thereof. The invention is more properly defined by the following claims. ##SPC1## ##SPC2## ##SPC3##