Consistent backups of data using a roll-back log

[0001] Businesses or other entities store their operational data (e.g., customer lists, financial transactions, business documents, business transactions, etc.) in logical volumes on memory devices such as hard disks. Unfortunately, the operational data is susceptible to corrupting events such as hardware failure, software failure, or user error. Hardware failures can range from the failure of a single hard disk to the destruction of an entire data center containing the hard disk, making some or all of the business's operational data unrecoverable. Software failures are procedural errors in an application that corrupt operational data. User errors include errors such as inadvertent deletion or overwriting of operational data that is later required. Failures and user errors often result in the loss or corruption of important operational data. The loss or corruption of operational data is particularly devastating to a business that relies heavily on electronic commerce.

[0002] Recognizing the importance of maintaining reliable operational data, businesses or other entities typically employ backup systems to protect critical operational data against corruption. Backup systems create backup or point-in-time (PIT) copies of operational data. A backup copy captures an image of the operational data at an instant in time so that if need be, everything that has happened to the operational data since that instant can be forgotten, and the state of operations (as reflected in the operational data) can be restored to that instant. Relatively simple backup programs can be used that interface with computer operating systems to backup data (e.g., files and directories stored on a hard disk) to magnetic tapes and to restore such data from the tapes onto the hard disk. Backup programs can run on standalone computer systems or on a server within a network of computer systems.

[0003] While it is a simple and fairly quick procedure to create a backup copy of an individual data object (image file, text file, etc.) the creation of a backup copy becomes more difficult as the amount of operational data increases. One method of creating a backup copy of a large volume of operational data is to copy the data from the hard disks that store the volume to one or more magnetic tapes. Once the backup has completed, the magnetic tapes are stored either locally or remotely. When a data corruption is detected in the operational data volume as a result of hardware failure, software failure, or user error, the volume is restored to its pre-corrupted data state using the backup copy.

[0004] Backup operations create backup copies that may be either full or incremental. A full backup copy means that all files in the data volume are copied regardless of how recently the files have been modified or whether a previous backup copy exists. An incremental backup means that only files of the volume that have changed since some previous event (e.g., a prior full backup or prior incremental backup) are copied. The backup window (the time needed to create the backup copy) for a full backup operation tends to be much larger when compared to the backup window for an incremental backup operation. For most applications, incremental backups are preferable at backup times since, in most cases, the number of files of the data volume that change between backups is very small compared to the number of files in the entire data volume and since the backup window is small. If backups are done daily or even more frequently, it is not uncommon for less than 1% of files of a volume to change between backups. An incremental backup in this case copies 1% of the data that a full backup would copy and uses 1% of the input/output (IO) resources between the hard disks and the backup magnetic tapes.

[0005] Incremental backup appears to be the preferred mode of protecting data. And so it is, until a full restore of all the files of the data volume is needed. A full restore from incremental backups requires starting with a restore using the newest full backup copy, followed by restores of all newer incremental backups. That can require a lot of magnetic tape handling performed by, for example, an automated robotic handler.

[0006] Restore from full backups is genuinely simpler and more reliable than restores from combinations of full and incremental backups. When recovering from individual user errors, the situation is just the opposite. Users tend to work with one small set of files for a period of days or weeks and then work with a different set. Accordingly, there is a high probability that a file erroneously corrupted by a user will have been used recently and therefore will be copied in one of the incremental backups. Since incremental backups contain a smaller fraction of data when compared to a full backup, the incremental backups can usually be searched much faster for the target file if a restore is required. From the individual user's standpoint, it is preferable to create many small incremental backups.

[0007] While backup is useful, it presents a number of disadvantages. The applications or database management systems that access volumes are required to be in “hot backup mode” before a backup operation starts on a data volume to ensure that the resulting back up copy is consistent with the data volume. Hot backup mode can impact the performance of applications or database management systems. For example, applications or database management systems cannot write new data to the volume until the backup operation has completed since the data contents of the volumes must remain unchanged during the backup window. Otherwise the resulting backup copy will not be consistent to a single point in time. A data storage log can be employed to store new volume data until the backup operation completes, and the volumes can be overwritten with the data stored in the storage log when the backup operation completes. However, these storage logs are complicated. Another problem is that applications often require that the application's backup method be used for backups. This limits the ability of an independent backup product to provide backup services without coordination with the application.

[0008] A system or method for creating a point-in-time (PIT) or backup copy of one or more data volumes using a roll-back log. The backup copy can be created without employing hot backup mode. In one embodiment, the method includes copying data blocks of a volume to a backup memory configured to store a backup copy of the volume, wherein the data blocks are sequentially copied until all of the data blocks of the volume are copied to the backup memory. A first request is generated to overwrite existing data of a first data block of the volume with new data before all of the data blocks of the volume are copied to the backup memory. Before overwriting the existing data and before all of the data blocks of the volume are copied to the backup memory, the existing data is copied to a first memory (e.g., the roll-back log). Before all of the data blocks of the volume are copied to the backup memory, the existing data is overwritten with the new data.

[0019] The use of the same reference symbols in different drawings indicates similar or identical items.

[0020] The present invention relates to a system or method for creating a point-in-time (PIT) or backup copy of one or more data volumes. Although the present invention will be described with reference to creating a backup copy on the volume level, the present invention may find application to creating a backup copy on the file system or database level. In one embodiment, the present invention allows the creation of backup copies without employing hot backup mode. For purposes of explanation, the present invention will be described with reference to creating a backup copy of one data volume V, it being understood that the present invention should not be limited thereto. Indeed, the present invention can be used to create a backup copy of several data volumes without the need of employing hot backup mode.

[0021] FIG. 1 illustrates relevant components of a data processing system 10 employing one embodiment of the present invention. It should be noted that the present invention can be employed in data processing systems other than that shown in FIG. 1. System 10 in FIG. 1 includes a computer system 12 coupled directly or indirectly to a memory system 16 that stores data volume V. Computer system 12 can be any computer or other data processing device. For example, computer system 12 can be a conventional personal computer, a server, a special purpose computer, or the like. For purposes of explanation, computer system 12 is presumed to take form in a server having one or more processors for executing instructions.

[0022] Server 12 generates transactions to directly or indirectly read data from or write data to volume V in response to executing an application program (not shown). A write data transaction is completed by overwriting existing data D_old in one or more data blocks of volume V with new data D_new. For ease of explanation only, it will be presumed that each write transaction, when completed, overwrites all existing data D_old in a single block of volume V with new data D_new, it being understood that the present invention should not be limited thereto. When a write transaction completes, there may be very little difference between existing data D_old and new data D_new.

[0023] Transaction requests to read or write data are transmitted directly or indirectly to memory system 16. Server 12 communicates directly or indirectly with memory system 16 via communication link 14. Communication link 14 may take form in a storage area network (SAN) having one or more components such as switches, hubs, routers (not shown), etc. System 10 may also include additional components (e.g., a database management system, a file system, a volume manager, etc.) in data communication with each other and with the application executing on server 12 and volume V, even though the additional components are not shown within FIG. 1.

[0024] Memory 16 may take form in one or more disk arrays or other storage devices, it being understood that the term memory system should not be limited thereto. For purposes of explanation, it will be presumed that memory system 16 takes form in a single disk array that stores logical data volume V. FIG. 1 illustrates a graphical representation of volume V. Volume V is an abstraction consisting of n_max data blocks that store or are configured to store data. While it is said that the data blocks of volume V store data, in reality, data is stored within respective physical memory blocks within disk array 16. The first five data blocks designated 1-5 of volume V in FIG. 1 are shown storing data designated A-E, respectively, at an initial time t₀. The last block n_max is shown storing data designated X.

[0025] FIG. 1 shows a backup storage system 24 in data communication with server 12. For purposes of explanation, it will be presumed that backup storage system 24 takes form in a robotic tape media handler with access to one or more magnetic data storage tapes. One or more tapes within backup storage system 24 store or are configured to store data of a backup copy V′ of volume V. Backup copy V′ is shown consisting of n_max data blocks that store or are configured to store a copy of data from n_max data blocks, respectively, of volume V. While it is said that the data blocks of backup copy V′ store data, in reality, data is stored within respective segments of one or more magnetic tapes. FIG. 1 shows backup copy V′ before volume V data is copied thereto.

[0026] FIG. 1 also show memory devices 18 and 20 coupled to and accessible by server 12. For purposes of explanation only, each of the memory devices 18 and 20 will take form in random access memory (RAM), it being understood that the present invention should not be limited thereto. Memory devices 18 and 20 may be configured as sequential logs. Data is read from sequential logs in the order in which the data is stored therein. Thus, the last data item stored in a sequential log is the first to be read out, and the first data item stored in a sequential log is the last to be read out.

[0027] Memory device 18 is designated as roll-back log while memory device 20 is designated as roll-forward log. Roll-back log 18 stores or is configured to store existing data D_old copied from volume V before D_old is overwritten with new data D_new of a corresponding write transaction. The roll-forward log 20 stores or is configured to store copies of the new data D_new. In one embodiment, the roll-forward log 20 may take form in a storage replication log (SRL). SRLs are typically used in systems for replicating data volumes.

[0028] Before or after existing data D_old is copied to roll-back log 18, in one embodiment, existing data D_old may be compressed according to any one of many well-known compression algorithms. Such compression would reduce the amount of storage space within roll-back log 18 needed to store data. Likewise, before or after new data D_new is copied to roll-forward log 20, new data D_new may also be compressed. For purposes of explanation, it would be presumed that data is stored in logs 18 and 20 in compressed format. Thus, before data can be read out and used from logs 18 and 20, the data must be decompressed.

[0029] Before existing data D_old in any block of volume V is overwritten with new data D_new, the existing data D_old is copied to roll-back log 18. New data D_new is copied to roll-forward log 20 either before existing data D_old is overwritten or after existing data D_old is overwritten. FIG. 2 describes this process in greater detail. For purposes of explanation, the process shown in FIG. 2 is implemented by server 12 executing instructions stored in memory (not shown). The process in FIG. 2 starts in step 40 when the application executing on server 12 generates a transaction to overwrite existing data D_old in block n of volume V with new data D_old. In step 42, before the existing data D_old in block n is overwritten with the new data D_new, the old data D_old is copied to roll-back log 18. In step 44, existing data D_old in block n is overwritten with new data D_new. In step 48, a copy of the new data D_new is stored in roll-forward log 20. It is noted that steps 44 and 48 may be reversed in order. Lastly, a tag Tm is generated. Tag Tm corresponds to the existing data D_old written to log 18 in step 42 and/or the new data D_new written to roll-forward log 20 in step 48. A new tag Tm is generated each time data is copied to logs 18 and/or 20. The tag number m is generated sequentially so that sequentially generated tags have sequential tag numbers. The application executing on server 12 may access data volume V while existing data D_old within volume V is copied to roll-back log 18 or while new data D_new is copied to roll-forward log 20.

[0030] Tag Tm may include a time stamp. The time stamp identifies the time when one of the steps shown within FIG. 2 completes. For example, the time stamp may be the time when the transaction is generated in step 40 by server 12. In the alternative, the time stamp may identify the time when new data D_new is copied to the roll-forward log 20. For purposes of explanation, it will be presumed that the time stamp is the time when D_old in block n of volume V is overwritten with new data D_new as shown in step 42.

[0031] Tag Tm will also be described as including an identification of the data block (i.e., block n) in volume V that is the target of the corresponding write transaction. In the embodiment where more than one data volume is accessed by the application of server 12, the tag Tm may include an identification of the volume that contains a target block of the corresponding write transaction. In the embodiment where more than one data block is the target of a corresponding write data transaction, the tag Tm may include an identification of the first data block and a number of consecutive blocks following the first data block where the new data D_new is to be written. In the embodiment where only a portion of existing data D_old in a data block is the target of the corresponding write data transaction, the tag Tm may include an identification of the data block, an offset from the beginning of the data block, and a data length of the new data D_new.

[0032] Tag Tm may also include information (e.g., a flag set to binary 1 or 0) indicating that data has or has not been stored within roll-back log 18 and/or roll-forward log 20 in compressed format. It is noted that other information may be included within the tags. The tag Tm may be stored alongside existing data D_old copied to log 18, alongside the new data D_new copied to log 20, or alongside both. In the alternative, Tag Tm may be stored in a separate tag table. Each tag Tm, however, is associated with a respective block of existing data D_old stored in roll-back log 18 and/or a respective block of new data D_new stored in roll-forward log 20.

[0033] FIG. 3a illustrates changes to the data contents of Volume V, roll-back log 18, and roll-forward log 20 after completing a first write transaction in accordance with the process shown in FIG. 2. More particularly, after time t₀ server 12 generates the first write data transaction to overwrite existing data (designated B) in block 2 of volume V with new data (designated B₁). Before existing data B of block 2 is overwritten, data B is copied to roll-back log 18 in accordance with step 42 of FIG. 2. Data B is stored in log 18 in compressed format. The existing data B in block 2 of volume V is then overwritten with new data B₁. The new data B₁ is copied to roll-forward log 20 either before or after existing data B is overwritten. Data B₁ is stored in roll-forward log 20 in compressed format. Lastly, server 12 generates tag T1 which it stores along with new data B₁ in roll-forward log 20 and with existing data B in roll-back log 18. Tag T1 includes the identification of the target block (e.g., block 2) of the first write transaction. Additionally, tag T1 includes a time stamp set to time t₁, the time when data B was overwritten in volume V. FIG. 3a shows state of volume V, roll-back log 18, and roll-forward log 20 after completion of the first write transaction in accordance with the process steps shown in FIG. 2.

[0034] After server 12 generates the first write transaction, server 12 generates a second write transaction to overwrite existing data (designated A) in block 1 of volume V with new data (designated A₁). Before existing data A of block 1 is overwritten, data A is copied to roll-back log 18. The existing data A in block 1 of volume V is then overwritten with the new data A₁ of the second write transaction. The new data A₁ is copied to roll-forward log 20 either before or after existing data A is overwritten. New data A₁ and existing data A are stored in compressed formats in logs 20 and 18, respectively. Lastly, server 12 generates a tag T2 which is stored along with new data A₁ in roll-forward log 20 and along with existing data A in roll-back log 18. Tag T2 includes the identification of the target block (e.g., block 1) of the second write transaction. Additionally, tag T2 includes a time stamp set to time t₂, the time when data A is overwritten in volume V. FIG. 3b shows the state of volume V at time t₂, and the state of roll-back log 18, and roll-forward log 20 after completion of the second write transaction in accordance with the process steps shown in FIG. 2.

[0035] Server 12 includes instructions stored in memory for creating a backup copy of volume V. It is noted that the backup process can be performed on a device external to server 12 in another embodiment. However, for purposes of explanation, it will be presumed that server 12 performs the backup operation. It will be presumed that the backup operation is scheduled to occur at time t_start, a time just after time t₂, the time when data A in block 1 of volume B is overwritten with the new data A₁ of the second write transaction.

[0036] FIG. 4 illustrates operational aspects performed by server 12 to create the backup copy V′ of volume V. More particularly, in step 50 server 12 sets a dummy variable n to 0. Thereafter in step 52, server increments n by 1. In step 54, server 12 copies the data contents of block n of volume V to block n of backup copy V′. In step 56, server 12 compares the current value of n with n_max, the number of the last data block in volume V. Steps 52 and 54 are repeated until n equals n_max. When n equals n_max in step 56, the process ends.

[0037] It is noted that data within volume V may be accessed via a read or write data transaction while data of volume V is being copied to backup copy V′ in accordance with the backup operation shown in FIG. 4. If volume V is modified via write data transaction before the backup operation completes, the backup copy V′ created by the process of FIG. 4 may not be an exact copy of data volume V as it existed at time t_start. FIGS. 5a-5c illustrate aspects of copying data from volume V to the backup storage system 24 as data within volume V is modified in accordance with write data transactions. After time t₂ and time t_start, server 12 generates a third write transaction to overwrite existing data (designated D) in block 4 of volume V with new data (designated D₁). Before existing data D of block 4 is overwritten, data D is copied to roll-back log 18. The existing data D in block 2 of volume V is then overwritten with new data D1 of the third write transaction. The new data D1 is copied to roll-forward log 20 either before or after existing data D is overwritten. Server 12 also generates a tag T3 which it stores along with the new data D1 in roll-forward log 20 and with existing data D in roll-back log 18. T3 includes the identification of the target block (e.g., block 4) of the third write transaction. Additionally, tag T3 includes a time stamp t₃ identifying the time when data D is overwritten in volume V with data D₁. Either before or after completion of the third write transaction, the data contents of blocks 1 and 2 of volume V are copied to blocks 1 and 2, respectively of backup copy V′ in accordance with the process of FIG. 4. FIG. 5a shows the state of volume V, roll-back log 18, and roll-forward log 20 after completion of the third write transaction at time t₃ in accordance with the steps shown in FIG. 2. Further, FIG. 5a also shows the state of backup copy V′ at time t₃.

[0038] Server 12 generates a fourth write transaction to overwrite existing data B₁ in block 2 of volume V with new data (designated B₂). Before existing data B₁ of block 2 is overwritten, B₁ is copied to roll-back log 18. The existing data B₁ in block 2 of volume V is then overwritten with the new data B₂ of the fourth write transaction. Data B₂ is copied to roll-forward log 20 either before or after existing data B₁ is overwritten in volume V. Server 12 generates a tag T4 which it stores along with new data B₂ in roll-forward log 20 and with a copy of existing data B₁ in roll-back log 18. Tag T4 includes the identification of the target block (e.g., block 2) of the fourth write transaction. Additionally, tag T4 includes a time stamp set to time t₄, the time when data B1 is overwritten in volume V. FIG. 5b shows the state of volume V, roll-back log 18, and roll-forward log 20 after completion of the fourth write transaction at time T₄ in accordance with the process steps shown in FIG. 2. Moreover, FIG. 5b shows the state of backup copy V′ after blocks 1-5 (and possibly more) of volume V have been copied to respective blocks within backup copy V′.

[0039] Lastly, in the illustrated example, server 12 generates a fifth write transaction to overwrite existing data (designated C) in block 3 of volume V with new data (designated C₁). Before existing data C of block 3 is overwritten, data C is copied to roll-backlog 18. Once data C has been copied to roll-back log 18, data within block 3 of volume V is overwritten with new data C₁. Server 12 generates a tag T5 which is stored along with new data C1 in roll-forward log 20 and with a copy of existing data C stored in roll-back log 18. T5 includes the identification of the target block (e.g., block 3) of the fifth write transaction. Additionally, T5 includes a time stamp set to time t₅, the time when data C is overwritten in volume V. FIG. 5c shows the contents of volume V, log 18, and log 20 after time t₅. Moreover, FIG. 5c shows that data has been copied from all of the data blocks in volume V to backup copy V′ in accordance with the process shown within FIG. 4. For the purposes of explanation, it will be presumed that the process shown in FIG. 4 ended at time t_end, it being understood that time t_end may have occurred after time t₅.

[0040] Because volume V was modified during the course of the backup operation described with reference to FIG. 4, backup copy V′ does not represent an exact copy of volume V at time t_start. For instance, data in block 4 of volume V is equal to D at time t_start. However, data D in block 4 was modified before the contents of block 4 was copied to backup copy V′. As such, block 4 within backup copy V′ contains data D₁. Roll-back log 18 can be used to incrementally restore backup copy V′ in reverse chronological order to equal the data state of volume V at the time t_start. FIG. 6 illustrates operational aspects of applying the contents of roll-back log 18 to backup copy V′ to incrementally restore backup copy V′. For ease of explanation, the process in FIG. 6 is implemented by server 12 executing instructions stored within memory (not shown). The process in FIG. 6 begins when a dummy variable c is set to m_end, where m_end is the tag number of the last tag generated before t_end (the time when the backup process shown in FIG. 4 ended). After step 60, server 12 compares the time stamp in tag Tc with time t_end, the time when the backup process shown within FIG. 4 ended. If the time stamp is later in time, server 12 decrements c by 1 and proceeds to step 66 wherein the time stamp of tag Tc is compared with time t_start, the time when the backup process shown within FIG. 4 was started. If the time stamp of tag Tc is later in time than t_start, data in backup copy V′ is overwritten with data corresponding to tag Tc and stored in the roll-back log 18. In step 66, the data overwritten in backup copy V′ is identified by the block number in tag Tc. If the time stamp in tag Tc is later in time when compared to time t_end in step 62, the process bypasses step 64 and proceeds immediately to step 66. Steps 68 and 64 are repeated in order until the time stamp in tag Tc is earlier in time than t_start in step 66. When this condition has been met, the process of FIG. 6 ends and backup copy V′ should equal the data state of volume V at time t_start. In other words, when the time stamp of tag Tc is earlier than t_start in step 66, backup copy V′ should be a consistent backup copy of volume V as it existed at time t_start. While contents of log 18 is being applied to backup copy V′ during the process of FIG. 6, volume V may be modified or accessed in accordance with read or write transactions generated by the application executing on server 12. For purposes of explanation, it will be presumed that volume V is not modified via a write data transaction after time t₅.

[0041] The process shown within FIG. 6 can be used to incrementally restore backup copy V′ in reverse chronological order from its data state shown within FIG. 5c to the data state of volume V shown within FIG. 3b. To illustrate, server 12 initially sets dummy variable c to 5, which is the value of m_end in the illustrated example. Accordingly, server 12 accesses tag T5. Tag T5 has a time stamp equal to time t₅, and tag T5 indicates that corresponding data C in log 18 was copied from block 3 of volume V. Since t₅ is presumed earlier in time when compared to t_end, server 12 skips step 64 and compares the time stamp of Tag 5 with t_start as shown in step 66. Since t₅ is later in time when compared to t_start, server 12 overwrites existing data in block 3 of backup copy V′ with data C from a roll-back log 18. It is noted that in this particular example, the existing data of block 3 of backup copy V′ initially contains data C. FIG. 7a shows the result of overwriting block 3 in backup copy V′ with data C in accordance with step 68 shown within FIG. 6.

[0042] Thereafter, server 12 decrements c in step 64 so that c now equals 4. Server than accesses the time stamp of tag T4, which has a time stamp t₄. Tag T4 also indicates that corresponding data B₁ in log 18 was copied to log 18 from block 2 of volume V. In step 66, server 12 compares t₄, the time stamp of tag T4 with t_start. Since t₄ is later than time t_start, server 12 overwrites existing data in block 2 of backup copy V′ with data B₁ from roll-back log 18. Again, in this particular example, the existing data of block 2 of backup copy V′ initially contains data B₁. FIG. 7b shows the result of overwriting block 2 in backup copy V′ with data B₁ in accordance with step 68 shown within FIG. 6.

[0043] Server 12 then accesses tag T3 after c is decremented to 3 in step 64. Tag T3 has a time stamp equal to t₃, and tag T3 indicates that corresponding data D in log 18 was copied to log 18 from block 4 of volume V. In step 66, server 12 compares t₃, the time stamp of T3 with t_start. Since t₃ is later than time t_start, server 12 overwrites existing data in block 4 of backup copy V′ with data D from roll-back log 18. FIG. 7c shows the result of overwriting block 4 in backup copy V′ with data D in accordance with step 68 shown within FIG. 6. At this point, backup copy V′ equals the data state volume V occupied at time t_start. Nonetheless, server 12 decrements c in step 64 so that c equals 2. Server 12 compares t₂, the time stamp of T2 with t_start. Since t₂ is earlier in time than time t_start, the restore process ends for the illustrated example, and backup copy V′ remains equal to the data state of volume V at time t_start.

[0044] After backup copy V′ has been restored in accordance with the process shown in FIG. 6, backup copy V′ optionally can be restored in forward chronological order to equal the data state of volume V at a point in time later than t_start. FIG. 8 illustrates operational aspects performed by server 12 for restoring backup copy V′ in forward chronological order. The process in FIG. 8 starts after a target time is selected. For purposes of explanation, it will be presumed that the target time will be set to t_end, the time at which the backup process shown in FIG. 4 ended. The process shown in FIG. 8 should be implemented after completion of the process shown in FIG. 6 so that the value of dummy variable c begins with the same value it had when the process of FIG. 6 ended.

[0045] The process in FIG. 8 begins with step 70 where server 12 increments c by 1. In step 72, the time stamp of tag Tc is compared with the target time (i.e., t_end). If the time stamp of Tc is earlier in time than t_end, the process proceeds to step 74 where server 12 overwrites existing data in backup copy V′ with data corresponding to tag Tc and stored in roll-forward log 20. Thereafter, server 12 increments c by 1 in step 76 and repeats step 72. Eventually, the time stamp of tag Tc will be later than the target time t_end in step 72, and the process will end. Once the process in FIG. 8 ends, backup copy V′ should have been restored to a data state equal the state of volume V at time t_end.

[0046] FIGS. 9a-9c illustrates the effects of applying the roll-forward log to the backup copy V′. To illustrate, server 12 initially increments c by 1 so that c is set to 3, remembering that c was set to 2 when the process ended in FIG. 6. With c set to 3, server 12 accesses tag T3, and in accordance with step 72 shown in FIG. 8, server 12 compares the time stamp t₃ of tag T3 with t_end. Given that the time stamp t₃ is earlier in time than t_end, the process proceeds to step 74 where server overwrites data D in block 4 of backup copy V′ with data D₁ of roll-forward log 20. It is noted that tag T3 indicates that existing data within block 4 of volume V was overwritten with data D₁ in roll-forward log 20 at time t₃. FIG. 9a illustrates the data state of backup copy V′ after data D is overwritten with data D₁ in step 74. C is incremented in step 76.

[0047] With c set to 4, server 12 accesses tag T4, and in accordance with step 72 shown in FIG. 8, server 12 compares the time stamp t₄ of tag T4 with t_end. Given that the time stamp t₄ is earlier in time than t_end, the process proceeds to step 74 where server overwrites data B1 in block 2 of backup copy V′ with data B₂ of roll-forward log 20. It is noted that tag T4 indicates that existing data within block 2 of volume V was overwritten with data B₂ in roll-forward log 20 at time t₄. FIG. 9b illustrates the data state of backup copy V′ after data B₁ is overwritten with data B₂ in step 74. C is incremented in step 76.

[0048] With c set to 5, server 12 accesses tag T5, and in accordance with step 72 shown in FIG. 8, server 12 compares the time stamp t₅ of tag T5 with t_end. Given that the time stamp t₅ is earlier in time than t_end, the process proceeds to step 74 where server overwrites data C in block 3 of backup copy V′ with data C₁ of roll-forward log 20. It is noted that tag T5 indicates that existing data within block 3 of volume V was overwritten with data C₁ in roll-forward log 20 at time t₅. FIG. 9c illustrates the data state of backup copy V′ after data C is overwritten with data C₁ in step 74. C is incremented in step 76. With c set to 6, server 12 accesses tag T6 (not shown), and in accordance with step 72 shown in FIG. 8, server 12 compares the time stamp of tag T6 with t_end. It will be presumed that the time stamp of tag T6 is later in time than t_end. As such, the process of FIG. 8 ends, and backup copy V′ should have a data state equal to the data state of volume V at time t_end.

[0049] Although the present invention has been described in connection with several embodiments, the invention is not intended to be limited to the specific forms set forth herein. On the contrary, it is intended to cover such alternatives, modifications, and equivalents as can be reasonably included within the scope of the invention as defined by the appended claims.