KR20130064319A - Hybrid hash index for storage device based on flash memory - Google Patents

Abstract

PURPOSE: A hybrid hash index for a flash memory-based storage device is provided to reduce the write and elimination operation of a flash memory by maximally delaying the division operation of an overflow bucket. CONSTITUTION: A flash memory(110) stores buckets and a control unit(130) manages the buckets. The control unit obtains a hash index about a key by using hash function and stores the key and an additional record including data in a bucket corresponding to the hash index. If a ratio of records to be merged is higher than a reference value, the control unit performs the merge operation of the bucket. If the ratio is smaller than the reference value, the control unit performs the merge and division operation of the bucket. [Reference numerals] (110) Flash memory; (120) Volatile memory; (130) Control unit

Description

HYBRID HASH INDEX FOR STORAGE DEVICE BASED ON FLASH MEMORY}

The present invention relates to a flash memory based storage device.

A flash memory based storage device is disclosed that manages buckets storing records using a hash index.

Various index structures, including hash indexes, may be used for disk-based storage devices such as hard disks.

When a disk-based extended hash index is used, the disk-based storage device performs a partition operation on the bucket in which the overflow occurs whenever a bucket overflow occurs. In addition, when a disk-based storage device deletes a record in which a key value is stored in a bucket, the disk-based storage device directly deletes the record.

Recently, flash memory has attracted attention as a storage medium to replace a hard disk for reasons of fast access speed, light weight, low power consumption, and shock resistance durability. Flash memory has different characteristics from disk-based storage in terms of its physical characteristics in terms of processing speed of operations such as read, write and erase, and in-place update operations.

When an existing hash index is used for a storage device based on the flash memory, the storage device based on the flash memory may perform the partitioning operation or the deleting operation when performing the partitioning operation or the delete operation on the record due to the bucket overflow. You must directly delete the record corresponding to. Therefore, a storage device based on flash memory cannot avoid an erase operation. In addition, in an update operation on a record, a storage device based on a flash memory must update a record stored in a bucket using a write operation and an erase operation.

Due to this feature, many problems may arise when the hash index implemented for the hard disk is used for the flash memory as it is.

In particular, frequent insertions, deletions, and updates of records can result in a large amount of logical in-situ updates to stored data. To physically handle such in situ updates, slow write and erase operations are performed in the NAND flash memory. Such slow write and erase operations can degrade the performance and durability of the NAND flash memory.

Korean Patent Publication No. 10-2010-0083976 (published Jul. 23, 2010) discloses a flash memory storage device using a hash table and a bucket for improving the write efficiency and lifetime of the flash memory storage device. Started.

One embodiment may provide a flash memory storage device that allocates an overflow bucket for a bucket in which an overflow has occurred and delays a partitioning operation on the bucket as much as possible, thereby reducing a write operation and an erase operation of the flash memory.

One embodiment may provide a flash memory storage device that reduces additional read and write operations by adaptively maintaining the number of overflow buckets for a bucket according to the ratio of update records and delete records in each bucket.

In one aspect, a flash memory for storing one or more buckets and a control unit for managing the one or more buckets, the control unit uses a hash function to obtain a hash index for the key, the one of the one or more buckets Storing an additional record containing the key and data in a bucket corresponding to a hash index,

If the maximum number of records are stored in the bucket, the controller performs a merge operation on the bucket if the ratio of records that can be merged among the records in the bucket is greater than or equal to the reference value, and the ratio is smaller than the reference value. A flash memory storage device is provided which performs a merge operation and a split operation on the bucket.

The control unit stores the additional record in the bucket as an insertion record when there is no record corresponding to the key in the corresponding bucket, and updates the additional record when there is a record corresponding to the key in the corresponding bucket. Can be stored in the corresponding bucket.

Records that can be removed by the merging include an update record and a delete record, wherein the update record indicates that data of the record corresponding to the key of the previously stored records of the additional record in the corresponding bucket is updated. The deletion record may indicate that the record corresponding to the key of the previously stored records of the additional record in the corresponding bucket has been deleted.

The controller may generate the space by additionally assigning an overflow bucket to the corresponding bucket when there is no space for storing the additional record in the corresponding bucket, and may store the additional record in the space.

The controller may perform a search operation by searching for a record corresponding to the key in a recently stored order in the bucket corresponding to the search key among the one or more buckets.

A flash memory storage device is provided that allocates an overflow bucket for an overflowed bucket to delay partitioning operations on the bucket as much as possible, thereby reducing the write and erase operations of the flash memory.

A flash memory storage device is provided that reduces additional read and write operations by adaptively maintaining the number of overflow buckets for a bucket in accordance with the ratio of update and delete records in each bucket.

1 is a structural diagram of a flash memory storage device according to an exemplary embodiment.
2 illustrates directories and buckets stored in a flash memory storage device according to an example.
3 illustrates an overflow processing method using an overflow bucket according to an example.
4 illustrates an update record and a delete record according to an example.
5 illustrates a result of merging and dividing a bucket according to an example.
6 illustrates a pseudo algorithm of an insert operation of a record according to an example.
7 illustrates the number of overflow buckets according to a merge criterion value according to an example.
8 illustrates index generation and search time according to the number of overflow buckets according to an example.
9 illustrates an operation count and a generation time in a hash index according to an example.
10 illustrates a search time of a hash index according to an example.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to or limited by the embodiments. Like reference symbols in the drawings denote like elements.

1 is a structural diagram of a flash memory storage device according to an exemplary embodiment.

The flash memory storage device 100 may include a flash memory 110, a volatile memory 120, and a controller 130.

The flash memory 110 may store one or more buckets, overflow buckets allocated to each bucket, a directory, and the like.

The volatile memory 120 may store or cache one or more buckets, overflow buckets and directories allocated to each bucket.

The controller 130 may manage one or more buckets, overflow buckets, directories, etc. to store data in the flash memory 110, and store one or more buckets, overflow buckets, directories, etc. in a volatile memory ( Caching within 120).

2 illustrates directories and buckets stored in a flash memory storage device according to an example.

In FIG. 2, directory 210 and two buckets 220 and 230 are shown. Entries in directory 210 may each represent one hash index value and may point to one bucket. An entry with a hash index value of "0" points to bucket number 220 and a hash index. A neck with a value of 1 indicates bucket 1 (230).

In FIG. 2, the directory 210 distinguishes buckets corresponding to keys using only one bit value. In other words, the hash index has a value composed of 1 bit. Thus, the global depth of directory 210 is one. In addition, bucket 0 and bucket 1 230 each correspond to one hash index value, and are divided by one bit among bits constituting the value of the hash index. That is, the local depth of each bucket 0 and bucket 1 230 is 1.

The controller 130 may obtain a hash index for the key using a hash function. The hash function may be a function that evenly distributes the keys to each of the one or more buckets. For example, the hash function may be a modular function that extracts a certain number of least significant bits among the bits constituting the key. In FIG. 2, the hash function may be a function that extracts the least significant bit of the key.

The bucket can store one or more records. In Figure 2, the bucket can store four records. In the bucket number 220, records having key values of 4, 12, and 10, respectively, were stored, and in the bucket number 230, records having key values of 1, 5, 21, and 16, respectively, were stored. The key value of the records in the bucket is shown, and the binary value of the key value is shown in parentheses.

The record is stored in one of the one or more buckets corresponding to the key of the record. Here, the bucket corresponding to the key of the record means the bucket corresponding to the hash index of the key. That is, when additionally storing the record, the controller 130 may store the additional record in the bucket corresponding to the hash index for the key among one or more buckets. Records stored may include keys and data.

By the hash function, the least significant one bit of the key value is extracted as the hash index. Thus, according to the binary representation of the key value, records with the key with the least significant bit 0 are stored in bucket 0, records with the key with the least significant bit 1 are stored in bucket 1 230. ,

3 illustrates an overflow processing method using an overflow bucket according to an example.

When the controller 130 performs an insert operation of a record, there may be no empty space in the bucket corresponding to the key value. For example, when the status of the buckets is as shown in FIG. 1, bucket 1 has already stored four records. Accordingly, when an additional record having a key value of 11 is stored, a bucket overflow occurs because there is no empty space in the bucket No. 230 corresponding to the key of the additionally stored record.

If a bucket overflow occurs because there is no space to store additional records in the bucket corresponding to the key, the controller 130 overflows to the corresponding bucket instead of immediately performing a split operation on the directory and one or more buckets. The space may be created by additionally assigning the bucket 310. The controller 130 may store the additional record in the space in the overflow bucket.

One or more buckets may each point to an overflow bucket assigned to it. A plurality of overflow buckets may be allocated to one bucket. The bucket and the plurality of overflow buckets allocated to the buckets may be connected in a line to each other through a pointer or list structure.

When the overflow bucket 310 is allocated to the bucket bucket 230, an additional record having a key value of 11 may be stored in the overflow bucket 310 of the bucket bucket 230. Subsequently, by an insert operation of the record, an additional record having a key value of 13 and an additional record having a key value of 29 may also be stored in the overflow bucket 310 of the first bucket 230 corresponding to the key.

The controller 130 may manage the bucket or the overflow only in the volatile memory 130 while there is a space to store the record in the bucket or the overflow bucket.

4 illustrates an update record and a delete record according to an example.

In FIG. 4, bucket 1 has been allocated two overflow buckets. As records are stored in the first bucket 230, a bucket overflow occurs, and the controller 130 transfers the first overflow bucket 410 and the second overflow bucket 420 to the first bucket 230. Assign.

Records stored in the bucket can be divided into insert records, delete records and update records. In FIG. 4, a first delete record 432, a second delete record 434, and a third delete record 436 are shown, and a first update record 442 is shown.

When the controller 130 inserts the additional record into the bucket, there may already be a record corresponding to the key (that is, a record having a key equal to the key of the additional record) in the bucket corresponding to the key of the record. Can be. If there is a record corresponding to the key in the bucket corresponding to the key of the additional record, the controller 130 may store the additional record in the bucket corresponding to the update record. That is, the update record may represent that the record corresponding to the key of the additional record is updated among the previously stored records of the additional record in the bucket corresponding to the key of the additional record. The first update record 442 may indicate that the data of the record 454 having the key value 13 in the bucket 230 is updated. Updated data of record 454 with key value 13 may be stored in first update record 442. Thus, after the update record is inserted, the valid record for key 13 becomes the first update record 442, not the record 454 with key value 13.

On the other hand, when the controller 130 inserts the additional record into the bucket, the record corresponding to the above key in the bucket corresponding to the key of the record (that is, the record having the same value as the key of the additional record). There can be no. If there is no record corresponding to the key in the bucket corresponding to the key of the additional record, the controller 130 may store the additional record as the insert record in the bucket.

That is, the controller 130 may process the insert operation or the update operation of the specific record by adding the specific record to the corresponding bucket. Therefore, even when data of a specific record is updated, the controller 130 does not need to physically update the record corresponding to the key of the specific record already stored in the bucket.

In addition, the controller 130 may process the insert operation of the record and the update operation of the record in the same manner, and the update record and the insert record may have the same form. The controller 130 may process the last stored record among the records having the same key stored in the bucket as a valid record for the key.

Due to the insertion of the update record as described above, the control unit 130 erases a block in the flash memory in which the record (or bucket) is stored when the data of the record having a specific key is to be updated. Can be avoided.

The controller 130 may delete the specific record in the bucket by adding the deletion record in the corresponding bucket. That is, the deletion record may indicate that the record corresponding to the key of the additional record among the records stored before the additional record in the bucket corresponding to the key of the additional record is deleted. For example, as shown in FIG. 4, when the record 452 with the key value 11 is stored in the bucket 1 230, the controller 130 may display a first deletion of the record 452 with the key value 11. Inserting the deletion record 432 into the bucket 1 may indicate that the record 452 having the key value 11 is deleted.

The controller 130 may generate the deletion record by negatively keying the key value of the record to be deleted. For example, the first deletion record 432 having a key value of −11 may be a deletion record indicating that the record 452 having a key value of 11 has been deleted.

Due to the addition of a deletion record as described above, when the control unit 130 wants to delete a record having a specific key, the control unit 130 avoids an operation of erasing a block in the flash memory in which the record (or bucket) is stored. Can be.

Then, when the record with the key value 11 is added again, the controller 130 may insert the record with the key value 11 into the bucket 230 again.

In accordance with the description of the update record and the insert record described above, there are five valid records in bucket 1 of FIG. 4, each having key values 19, 13, 5, 15 or 21.

According to the description of the update record and the insert record described above, it can be seen that the record stored at the end of the records having the same (or the same absolute value) key stored in the bucket is a valid record for the key. Therefore, when a record having a search key is searched, the controller 130 may perform a search operation by searching for the records corresponding to the key in the order of recently stored in the bucket corresponding to the search key among one or more buckets. have.

5 illustrates a result of merging and dividing a bucket according to an example.

The number of records that can be stored in one bucket, including records stored in the overflow bucket, can be limited. For example, the number of records that can be stored in a bucket without an overflow bucket being allocated is four, the number of records that can be stored in one overflow bucket is four, and the overflow can be allocated to one bucket. If the maximum number of buckets is two, the maximum number of records that can be stored in one bucket is twelve.

When a maximum number of records are stored in the bucket, the controller 130 may perform a merge operation and a split operation on the records in the bucket.

The merge operation may mean reducing the number of records in the bucket by merging the records in the bucket. Here, the record that can be merged may include an update record and a delete record. For example, when a merge operation is applied to a bucket, the controller 130 may delete a specific record in the bucket and a delete record corresponding to the specific record. In addition, when the merge operation is applied to the bucket, the controller 130 may delete the specific record and leave only the update record among the specific record in the bucket and the update record corresponding to the specific record. When there are a plurality of update records for a specific record of the honey, the controller 130 may leave only the last update record among the specific record and the plurality of update records.

The partitioning operation may mean reducing the number of records stored in each bucket by dividing and storing records in one bucket into two buckets. The controller 130 may divide one bucket into two buckets through a split operation. The controller 130 may obtain a hash index by applying a hash function to the key of each record for each record stored in the divided bucket, and among the two buckets newly created according to the hash index of each record. Each record can be stored in one bucket.

The controller 130 may set the value of the local depth of each of the buckets generated by the division to be greater than the value of the local depth of the divided bucket. In addition, when the value of the local depth of the divided bucket is equal to the global depth, the controller 130 may increase the global depth by 1 so that the entry of the directory 210 may point to the bucket created by the division. 210 may be divided.

When the directory 210 is not divided, the controller 130 causes two items in the directory 210 that pointed to the bucket that is the target of the splitting to point to one bucket among the buckets created by the splitting.

When the directory 210 is divided, the controller 130 allows each item in the directory to point to an existing bucket other than the divided bucket or a new bucket created by the split.

When the maximum number of records is stored in a particular bucket, if the ratio of records that can be merged among the stored records is above a certain reference value, the number of records may be sufficiently reduced by merging, and the bucket may not need to be divided. . Therefore, the controller 130 may store only the maximum number of records in the bucket, and if the ratio of records that can be merged among the records in the bucket is greater than or equal to the reference value, the controller 130 may perform only a merge operation on the bucket.

On the other hand, if the ratio is less than a certain reference value, many records remain in the particular bucket even after merging. Accordingly, the controller 130 stores the maximum number of records in the bucket, and if the ratio of records that can be merged among the records in the bucket is smaller than the reference value, the controller 130 first performs a merge operation on the bucket, and then merges the bucket. You can perform partitioning operations on buckets. The controller 130 may reconstruct the buckets generated by the division into valid records remaining after the merge.

Here, the ratio may be a ratio of the deletion record and the update record to the maximum number of records that can be stored in the bucket. For example, 12 records are stored in the first bucket 230 of FIG. 3, and there are three double deletion records and one update record. Therefore, in the bucket 230 of FIG. 3, the ratio may be 4/12 (= 1/3). Alternatively, the ratio may be the number of records to be removed after merging with respect to the maximum number of records that can be stored in the bucket. The delete record removes both the delete record itself and the deleted record. The update record removes records (including update records) with the same key previously stored. Therefore, in the bucket 230 of FIG. 3, the ratio may be 7/12. The reference value may be expressed by the merge reference value α.

The controller 130 may adaptively determine the merge reference value α based on the characteristics of the flash memory or the characteristics of the operations performed by the flash memory storage device 100.

By merging and splitting as described above, the delete record and the update record can be removed in the bucket. In addition, an erase operation may be delayed for a block in which records of a bucket are stored until a merge or split occurs, so that the performance of the flash memory storage device 100 may be improved.

For example, when the value of α is 0.5 and the ratio is 7/12, the controller 130 may merge the bucket 230 of FIG. 4, and the valid records remaining after the merge (ie, the key value) The first bucket 510 and the second bucket 520 of FIG. 5 may be generated by dividing the records of 13, 5, 21, 19, or 15). Due to this split, the depth of directory 210 increases to two. Directory 210 uses two bit values to distinguish the bucket corresponding to the key. In other words, the hash index has a value composed of 2 bits. Each bucket is divided by one or two bits of the bits constituting the value of the hash index, corresponding to one hash index value.

In FIG. 5, bucket 0 corresponds to hash indices “00” and “10”. That is, a record having a key having a value of "0" of the least significant 1 bit of the hash index is stored in the bucket number 210. The first bucket 510 corresponds to the hash index "01". That is, a record having a key whose value of the least significant two bits of the hash index is "01" is stored in the first bucket 510. Bucket number 520 corresponds to hash index " 11 ". That is, a record having a key whose value of the least significant two bits of the hash index is "11" is stored in bucket number 510. The local depth of bucket 0 is 220, and the depth of bucket 1 510 and bucket 520 is 2. Different levels of hash function may be used for each bucket having a different local depth. That is, compared to the first hash function used for the first bucket 510, the second hash function used for the zero bucket 220 may be a hash function one step below.

For a search operation, the controller 130 may convert the search key value into a hash value representing a hash index using a hash function. The controller 130 may find a bucket (or bucket address) corresponding to the key value through the directory 210. The controller 130 may search for the record including the key having the same value as the search key value through the address of the bucket. As described above, since the delete record and the update record are inserted into the bucket instead of the erase operation on the record, the control unit 130 stores the latest record in the bucket from the location where the oldest record is stored (i.e., in the reverse direction). You can check the keys of the records.

The hash index used by the same method as described above is referred to as hybrid hash index hereinafter.

6 illustrates a pseudo algorithm of an insert operation of a record according to an example.

The first row may indicate obtaining a hash value by using the key value of the hash function.

Rows 2 through 5 may indicate inserting a record into the bucket corresponding to the hash value when the full depth and the use depth of the bucket corresponding to the hash value are the same.

Rows 6 to 11 obtain a new hash value for the key by using the hash function one step below when the full depth and the bucket depth corresponding to the hash value are not the same, and then apply the new hash value. Inserting a record into a corresponding bucket.

Rows 12 to 21 measure the ratio of delete records and update records in the bucket when the maximum number of records are inserted in the bucket, and perform the merge and split operations when the measured ratio is equal to or greater than the merge criterion value α. Can be represented. At this time, if the measured ratio is smaller than the merge criterion value α, the bucket can be reconstructed with valid records obtained by the merge operation. The insert operation on the reconstructed bucket can be performed again until the reconstructed bucket has stored the maximum number of records.

7 through 10 illustrate simulations of a flash memory storage device, according to an exemplary embodiment. In the experiment described with reference to FIGS. 7 to 10, one block is composed of 32 pages. One block can store 32 key values (ie records). To measure the creation time of the index, the number of read, write and erase operations of the flash memory when 10,000 records were inserted were measured. To calculate the final generation time, the flash memory speed was set to 25 μs, 200 μs and 1.5 ms for read, write and erase operations, respectively.

7 illustrates the number of overflow buckets according to a merge criterion value according to an example.

The first graph 710 represents an index generation time for each merge criterion value when the maximum number of overflow buckets allocated to the bucket is two.

The second graph 720 shows the index generation time for each merge criterion value when the maximum number of overflow buckets allocated to the bucket is eight.

The index creation time was measured for each merge criterion value as the merge criterion values 盼 were set differently from 0 to 0.9. When keys (or records) are inserted up to the maximum overflow bucket of each bucket, it is determined whether to perform the post-merger division operation or the merge operation only according to the merge criterion value. The first graph 710 and the second graph 720 represent the generation time of the hash index according to the merge criterion value for each of the traces having the update and delete rates of 20%, 50%, and 80%, respectively.

According to the graphs shown, when the merge criterion value is 0 or more and 0.4 or less, the index generation time decreases as the merge criterion value increases, but the width of the decrease in creation time decreases as the merge criterion value becomes 0.4 or more. Hereinafter, the merge reference value is set to 0.4 in order to measure the number of operations and the overall performance of the flash memory according to the update and delete rates.

8 illustrates index generation and search time according to the number of overflow buckets according to an example.

When the maximum number of overflow buckets that can be allocated to a bucket was set from 0 to 12, the time for generating a hash index, the average time for searching an inserted key, and the bucket utilization were measured. Traces with update and delete rates of 20%, 50%, and 80%, respectively, were used, and the merge baseline was set to 0.4.

The first graph 810 shows an index generation time according to the number of overflow buckets. The first graph 810 shows the fastest index creation time when the number of overflow buckets is 4 with the update and delete ratio of 20%, and the index creation time gradually increases as the number of overflow buckets increases. It gradually decreases. This change in index creation time occurs because the number of partition or merge operations that cause an erase operation is significantly reduced when the overflow bucket has a sufficient number of overflow buckets.

For traces with update and delete rates of 50% and 80%, the first graph 810 shows the fastest index creation time when the number of overflow buckets is 4 or 8; otherwise, overflow It looks similar when the number of buckets is 4 or 8.

The second graph 820 represents an average search time for each trace according to the number of overflow buckets. In the second graph 820, as the number of overflow buckets increases, search performance decreases linearly. This degradation occurs because all records in the overflow bucket allocated to the search operation must be sequentially searched.

The third graph 830 shows bucket utilization for each trace according to the number of overflow buckets. The third graph 830 generally shows that when the number of overflow buckets is small (i.e. when the number of overflow buckets is 0 or 1), the search performance is improved, but the bucket utilization is 50 to 60%. It shows a marked drop. The low bucket utilization rate may be a disadvantage in an environment where computing resources are limited. Bucket utilization can increase as the number of merge operations occurs more frequently. The increase in bucket utilization as the number of merge operations occurs frequently means that the overall depth of the hash index does not increase when a merge operation occurs, so it records only existing buckets without additional bucket allocation. This is because the insertion is made. Increasing the number of overflow buckets increases the probability that a merge operation will occur rather than a split operation. Therefore, as shown in the third graph 830, as the number of overflow buckets increases, the bucket utilization rate increases.

According to the above experiment, when the maximum number of overflow buckets for each bucket is 4 or 8, an optimal index generation time can be derived. However, in terms of search time, when the number of overflow buckets is eight, on average, about two times as long as when the number of overflow buckets is four. Therefore, in consideration of both index generation and search performance, the maximum number of overflow buckets for each bucket is set to four.

9 illustrates an operation count and a generation time in a hash index according to an example.

In FIG. 9, for each of the extended hash index and the hybrid hash index of the present invention used on the flash memory, the number of read operations, write operations, and erase operations generated by the index generation is measured, respectively. The creation time of the hash indexes is compared.

By inserting 10,000 random key records according to the update and delete rate from 0% to 90%, the number of read, write and erase operations by the insert was measured and the final generation time was calculated. . As described above, the number of overflow buckets is set to 4 and the merge criterion value is set to 0.4.

Extended hash indexes perform partitioning operations whenever an overflow occurs. Therefore, when the extended hash index is used, more read operations, write operations, and erase operations may occur as compared with the use of the hybrid hash index which delays the split operation. In addition, the higher the update and delete ratio, the more in-place update may occur. Extended hash indexes do not avoid write-before-write operations whenever an in-place update occurs. Thus, write operations and erase operations may additionally occur. However, if there are few keys that have been updated and deleted (e.g., 0% to 10%), even in hybrid hash indexes, partitioning operations for overflow buckets may occur collectively in each bucket. Thus, if there are few keys (or records) updated and deleted, the number of write and erase operations generated by the extended hash index may be less than when the hybrid hash index is used.

The first graph 910, the second graph 920, the third graph 930, and the fourth graph 940 each generate a generation period, a number of read and draw operations, a number of write operations, and an erase occurring for each hash index. Indicates the number of operations. The graphs shown linearly increase the creation time of the extended hash index and the number of each of the operations as the update rate of the key (or record) increases. On the other hand, in the hybrid hash index, since the update record and the delete record are inserted into the record of the bucket as they are, and then merged, the pre-write-erase operation accompanying the in-place update and the in-place update can be avoided. Thus, the number of read, write and erase operations of the hybrid hash index may be less than the total number of operations of the extended hash index as a whole.

10 illustrates a search time of a hash index according to an example.

The time to retrieve all the key values inserted into the generated hash index is measured. The number of overflow buckets was set to 4 and the merge criterion was set to 0.4.

Graph 1010 represents the average search time of each hash index according to the update and delete rates. Extended hash indexes perform partitioning operations whenever an overflow occurs. Therefore, the higher the update and delete ratio, the shorter the search time.

Since the merge criterion value is 0.4, when the hybrid hash index is used, if the update and delete ratio of each trace is 0% to 30%, the merge operation and the split operation may be performed more frequently than when only the merge operation is performed. . In this case, the higher the update and delete ratio of the trace, the shorter the search time. However, because buckets are additionally allocated by the partitioning operation, the page utilization can be further reduced.

On the other hand, if the update and delete ratio is 40% or more, only the merge operation may be performed more often without the split operation, and the page utilization may increase. Thus, search performance may be temporarily degraded. However, if the update and delete rate is higher, such as 50% to 90%, since the number of valid records in the bucket is smaller, high page utilization and fast search performance can be exhibited at the same time.

In search performance, the search performance of the extended hash index as a whole may be better than the search performance of the hybrid hash index. However, as shown in graph 1010, the search speed difference between the two indices is only 1 ms. In addition, in the flash memory, since the speed of the read operation is very fast compared to the speed of the write operation and the erase operation, the time increased by the additional read operation during the search operation may not significantly affect the overall performance of the search operation. .

The method according to one embodiment may be implemented in the form of program instructions that may be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, the present invention has been described by way of limited embodiments and drawings, but the present invention is not limited to the above embodiments, and those skilled in the art to which the present invention pertains various modifications and variations from such descriptions. This is possible.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the equivalents of the claims, as well as the claims.

100: flash memory storage device
110: flash memory
120: volatile memory
130:
210: directory
220: bucket 0
230: bucket 1

Claims (5)

Flash memory for storing one or more buckets; And
Control unit for managing the one or more buckets
Lt; / RTI >
The control unit obtains a hash index for a key using a hash function, stores an additional record including the key and data in a bucket corresponding to the hash index among the one or more buckets,
If the maximum number of records are stored in the bucket, the controller performs a merge operation on the bucket if the ratio of records that can be merged among the records in the bucket is greater than or equal to the reference value, and the ratio is smaller than the reference value. And merge operation and division operation on the bucket. The method of claim 1,
The control unit stores the additional record in the bucket as an insertion record when there is no record corresponding to the key in the corresponding bucket, and updates the additional record when there is a record corresponding to the key in the corresponding bucket. Flash memory storage in the corresponding bucket. The method of claim 1,
Records that can be removed by the merging include an update record and a delete record, wherein the update record indicates that data of the record corresponding to the key of the previously stored records of the additional record in the corresponding bucket is updated. And the deletion record indicates that the record corresponding to the key of the previously stored records of the additional record in the corresponding bucket is deleted. The method of claim 1,
And the controller generates the space by additionally assigning an overflow bucket to the corresponding bucket when there is no space to store the additional record in the corresponding bucket, and stores the additional record in the space. . The method of claim 1,
And the control unit performs a search operation by searching for a record corresponding to the key in a recently stored order in a bucket corresponding to a search key among the one or more buckets.

Images