Fast write operations

• US 4,916,605 A
• Filed: 08/31/1987
• Issued: 04/10/1990
• Est. Priority Date: 03/27/1984
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Howard, John H., Canon, Michael D., Easton, Malcolm C., Beardsley, Brent C., Hartung, Michael H., Vosacek, Robert H.
• Primary Examiner(s): Clark, David L.
• Assistant Examiner(s): Jaffe, Michael A.
• Application Number:
  US07/091,406
• Time in Patent Office: 953 Days
• Field of Search: 364/200 MS File, 364/ 900 MS File
• US Class Current: 711/162
• CPC Class Codes: G06F 11/1471 involving logging of persistent data for recovery
• CPC Class Codes: G06F 11/1666 where the redundant component is memory or memory area
• CPC Class Codes: G06F 11/20 using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
• CPC Class Codes: G06F 12/0804 with main memory updating G06F12/0806 takes precedence
• CPC Class Codes: G06F 12/0866 for peripheral storage systems, e.g. disk cache
• CPC Class Codes: G06F 2212/312 In storage controller

Abstract

A technique is described for performing a fast write operation. A host write request, which would normally be serviced by an immediate physical write to a data storage device, is instead written to cache and nonvolatile storage in the data storage device controller. Then, the controller signals the host that the write operation is complete and does not update the physical data storage device until later. A journal log is also used to provide recovery capability in the event of system failure. This technique provides high performance for the units'"'"' operation while assuring integrity by keeping two copies of the write operation until the physical update transpires.

Claims

1. In a data processing system having a host system, a controller attached to said host system, and a plurality of data storage devices connected to said controller, said controller including a cache, a nonvolatile storage, and a memory containing a status table, said controller comprising attaching means for said plurality of data storage devices, each of said data storage devices having a device status track having a device identification number and controller identification number recorded thereon, said status table in said memory comprising status information on said nonvolatile storage, said cache, a fast write condition indicating a delayed writing of data stored in said nonvolatile storage to one of said data storage devices when active, and a device identification number and a controller identification number of each of said data storage devices, a method of performing a fast write operation comprising the steps of:

accepting a write request from said host for writing data to one of said data storage devices;

checking said status table to determine if said nonvolatile storage is available, if said cache is active, and if said fast write condition is active;

if said nonvolatile storage is available, said cache is active and said fast write condition is active, then storing duplicate copies of said data in said nonvolatile storage and said cache;

signalling said host that a write operation completed successfully as soon as said duplicate copies of said data are successfully written to said nonvolatile storage and said cache; and

writing said data stored in said nonvolatile storage to one of said data storage devices at a later time;

but if either said nonvolatile storage is unavailable or said fast write condition is inactive, then immediately destaging data stored in said cache to one of said data storage devices; and

in the case where said cache is inactive, then immediately destaging data stored in said nonvolatile storage to one of said data storage devices.

2. The method of performing a fast write operation as recited in claim 1, further comprising the step of preventing an initial memory load of a data storage device from successfully completing unless said controller identification number and said device identification number in said status table of said controller match said device and controller identification numbers on said device status track of said data storage device.

3. The method of performing a fast write operation as recited in claim 1, wherein said data storage device has a standard format with a record zero and a predictable write operation is performed, comprising the steps of:

writing a track of information into said cache and said nonvolatile storage; and

at a later time, orienting said data storage device to said record zero and destaging said track of information from said cache to said data storage device beginning at said record zero.

4. The method of performing a fast write operation as recited in claim 3, further comprising the step of responding to a power loss by destaging said track of information from said nonvolatile store to said data storage device after power is restored.

5. The method of performing a fast write operation as recited in claim 1 wherein said nonvolatile storage comprises a nonvolatile buffer and a circular journal recorded on a nonvolatile medium, said nonvolatile buffer having a capacity less than the capacity of said cache, said method further comprising the steps of:

periodically transferring records in said nonvolatile buffer onto a partition of said circular journal; and

periodically transferring selected entries from one partition to another partition of said circular journal after records in said nonvolatile storage have been written to a data storage device, whereby the most recent copy of each changed record in said cache always appears in either said nonvolatile buffer or said circular journal.

6. The method of performing a fast write operation as recited in claim 5 wherein each record in cache includes a status field, further comprising the steps of:

if a record in cache matches a corresponding record on a data storage device, setting said status field to zero;

but

if a record in cache has been changed but not yet written to a data storage device, setting said status field to a value which points to a location in said circular journal to which the record is to be written.

7. A data processing system for performing a fast write operation comprising:

a host system containing a plurality of channels;

a controller attached to at least one of said channels and comprising a cache, a nonvolatile storage, and a memory containing a status table;

a plurality of data storage devices attached to said controller, each of said data storage devices comprising a plurality of tracks having a plurality of records recorded thereon and a device status track having a device identification number and a controller identification number recorded thereon;

said status table storing status information on said nonvolatile storage, said cache, a fast write condition indicating a delayed writing of data stored in said nonvolatile storage to one of said data storage devices when active, and a device identification number and a controller identification number for each of said storage devices;

means for accepting a write request from said host to write data to one of said data storage devices;

means for checking said status table to determine if said nonvolatile storage is available, if said cache is active, and if said fast write condition is active;

means responsive to said checking means for storing duplicate copies of said data in said nonvolatile storage and said cache when said nonvolatile storage is available, said cache is active and said fast write condition is active;

means for signalling said host that a write operation completed successfully as soon as said duplicate copies of said data are written to said nonvolatile storage and said cache;

means for writing said data stored in said nonvolatile storage on one of said data store devices at a later time;

means responsive to said checking means when either said nonvolatile storage is unavailable or said fast write condition is inactive for immediately destaging data stored in said cache to one of said data storage devices; and

means responsive to said checking means when said cache is inactive for immediately destaging data stored in said nonvolatile storage to one of said data storage devices.

8. The data processing system as recited in claim 7, further comprising means for assuring integrity of data storage devices after failures by comparing said device identification number in said status table with said device identification number and said controller identification number on said device status track of said data storage device and disabling operation of said data processing system if the device identification number and controller identification number do not match.

9. The data processing system as recited in claim 7 wherein said nonvolatile storage comprises a nonvolatile buffer and a circular journal recorded on a nonvolatile medium, said data processing system further comprising:

means for periodically transferring data in said nonvolatile buffer onto a partition of said circular journal; and

means for periodically transferring selected entries from one partition to another partition of said circular journal after records in said nonvolatile storage have been written on a data storage device.

10. In a staged storage system wherein at least one processor accesses a volatile primary store in which data are destaged to a secondary store under a Least Recently Used (LRU) replacement algorithm, and data changes are made in records that are written initially to the primary store and thereafter posted to the secondary store only upon destaging, and data is formatted into records and pages, a method for protecting the primary store contents comprising the steps of:

(a) responsive to each processor access of the primary store for altering a record, writing a copy of each changed record including a status field and time stamp into a nonvolatile buffer concurrent with updating of an accessed record in the primary store, each changed record and its copy having its status field set to a same value;

(b) periodically copying changed records with status fields and time stamps from the buffer onto a circular journal stored on a nonvolatile medium, the journal having finite storage space and being partitioned into k consecutively addressable equal-sized extents, the status field being indicative as to which of the k journal partitions the record belongs, and the time stamp showing an order in which entries were written; and

(c) during an interval within which changed records are being written into the buffer for eventual copying onto a journal partition i, selecting records in the primary store that have not been destaged and whose status fields indicate their association with the next journal partition in sequence following i, and writing said selected records into journal partition i.

11. A method according to claim 10, wherein the journal is partitioned into k consecutively addressable equal-sized extents and data is formatted into pages and records, each record having a status field indicative of whether the record has had either no changes thereto or in which journal partition changes of the record were written and each page having a status field indicative of whether there may exist stale records within the page.

12. A method according to claim 11, wherein the status field of each page is used to control the selection and rewriting of changed records that are not stale and which still reside in the primary store.

13. A method according to claim 10, wherein during an interval within which changed records are being written to the journal and upon the buffer being filled, destaging additional records to the secondary store which would otherwise be destined to be written into the journal.

14. A method according to claim 13 wherein if L is the number of bytes in each journal partition, N is the number of bytes in the primary store, B is the number of bytes in the buffer, M is a maximum record size in bytes, and H is a threshold in bytes of an amount of journal partition determinative when to complete scanning of the primary store for selecting stale records, then

15. A method according to claim 10, wherein each page includes a status field indicative of whether there may exist stale changed records within the page and further wherein the page status field is used to control scanning and selection of prospectively stale records in the volatile primary store for determining which corresponding records in the journal are not stale and should be rewritten.