The Berkeley DB access method implementation unavoidably interacts with each application's data set, locking requirements and data access patterns. For this reason, one access method may result in dramatically better performance for an application than another one. Applications whose data could be stored using more than one access method may want to benchmark their performance using the different candidates.
One of the strengths of Berkeley DB is that it provides multiple access methods with nearly identical interfaces to the different access methods. This means that it is simple to modify an application to use a different access method. Applications can easily benchmark the different Berkeley DB access methods against each other for their particular data set and access pattern.
Most applications choose between using the Btree or Heap access methods, between Btree or Hash, or between Queue and Recno, because each of these pairs offer similar functionality.
Most applications use Btree because it performs well for most general-purpose database workloads. But there are circumstances where Heap is the better choice. This section describes the differences between the two access methods so that you can better understand when Heap might be the superior choice for your application.
Before continuing, it is helpful to have a high level understanding of the operating differences between Btree and Heap.
The Heap access method was developed for use in systems with constrained disk space (such as an embedded system). Because of the way it reuses page space, for some workloads it can be much better than Btree on disk space usage because it will not grow the on-disk database file as fast as Btree. Of course, this assumes that your application is characterized by a roughly equal number of record creations and deletions.
Also, Heap can actively control the space used by the database with the use of the DB->set_heapsize() method. When the limit specified by that method is reached, no additional pages will be allocated and existing pages will be aggressively searched for free space. Also records in the heap can be split to fill space on two or more pages.
Btree and Heap are fundamentally different because of the way that you access records in them. In Btree, you access a record by using the record's key. This lookup occurs fairly quickly because Btree places records in the database according to a pre-defined sorting order. Your application is responsible for constructing the key, which means that it is relatively easy for your application to know what key is in use by any given record.
Conversely, Heap accesses records based on their offset location within the database. You retrieve a record in a Heap database using the record's Record ID (RID), which is created when the record is added to the database. The RID is created for you; you cannot specify this yourself. Because the RID is created for you, your application does not have control over the key value. For this reason, retrieval operations for a Heap database are usually performed using secondary databases. You can then use this secondary index to retrieve records stored in your Heap database.
Note that an application's data access requirements grow complex, Btree databases also frequently require secondary databases. So at a certain level of complexity you will be using secondary databases regardless of the access method that you choose.
Secondary databases are described in Secondary indexes.
When Btree creates a new record, it places the record on the database page that is appropriate for the sorting order in use by the database. If Btree can not find a page to put the new record on, it locates a page that is in the proper location for the new record, splits it so that the existing records are divided between the two pages, and then adds the new record to the appropriate page.
On deletion, Btree simply removes the deleted record from whatever page it is stored on. This leaves some amount of unused space ("holes") on the page. Only new records that sort to this page can fill that space. However, once a page is completely empty, it can be reused to hold records with a different sort value.
In order to reclaim unused disk space, you must run the DB->compact() method, which attempts to fill holes in existing pages by moving records from other pages. If it is successful in moving enough records, it might be left with entire pages that have no data on them. In this event, the unused pages might be removed from the database (depending on the flags that you provide to DB->compact()), which causes the database file to be reduced in size.
Both tree searching and page compaction are relatively expensive operations. Heap avoids these operations, and so is able to perform better under some circumstances.
Heap does not care about record order. When a record is created in a Heap database, it is placed on the first page that has space to store the record. No sorting is involved, and so the overhead from record sorting is removed.
On deletion, both Btree and Heap free space within a page when a record is deleted. However, unlike Btree, Heap has no compaction operation, nor does it have to wait for a record with the proper sort order to fill a hole on a page. Instead, Heap simply reuses empty page space whenever any record is added that will fit into the space.
When considering Heap, be aware that this access method does not support the full range of cursor operations that Btree does.
On sequential cursor scans of the database, the retrieval order of the records is not predictable for Heap because the records are not sorted. Btree, of course, sorts its records so the retrieval order is predictable.
When using a Heap database, you cannot
create new records using a cursor. Also, this
means that Heap does not support the DBC->put()
DB_AFTER and
DB_BEFORE flags. You
can, however, update existing records using a
cursor.
For concurrent applications, iterating through the records in a Heap database is not recommended due to performance considerations. This is because there is a good chance that there are a lot of empty pages in the database if you have a concurrent application.
For a Heap database, entire regions are locked when a lock is acquired for a database page. If there is then contention for that region, and a new database page needs to be added, then Berkeley DB simply creates a whole new region. The locked region is then padded with empty pages in order to reach the new region.
The result is that if the last used page in a region is 10, and a new region is created at page 100, then there are empty pages from 11 to 99. If you are iterating with a cursor, then all those empty pages must be examined by the cursor before it can reach the data at page 100.
Ultimately, you can only determine which access method is superior for your application through performance testing using both access methods. To be effective, this performance testing must use a production-equivalent workload.
That said, there are a few times when you absolutely must use Btree:
If you want to use bulk put and get operations.
If having your database clustered on sort order is important to you.
If you want to be able to create records using cursors.
If you have multiple threads/processes simultaneously creating new records, and you want to be able to efficiently iterate over those records using a cursor.
But beyond those limitations, there are some application characteristics that should cause you to suspect that Heap will work better for your application than Btree. They are:
Your application will run in an environment with constrained resources and you want to set a hard limit on the size of the database file.
You want to limit the disk space growth of your database file, and your application performs a roughly equivalent number of record creations and deletions.
Inserts into a Btree require sorting the new record onto its proper page. This operation can require multiple page reads. A Heap database can simply reuse whatever empty page space it can find in the cache. Insert-intensive applications will typically find that Heap is much more efficient than Btree, especially as the size of the database increases.
The Hash and Btree access methods should be used when logical record numbers are not the primary key used for data access. (If logical record numbers are a secondary key used for data access, the Btree access method is a possible choice, as it supports simultaneous access by a key and a record number.)
Keys in Btrees are stored in sorted order and the relationship between them is defined by that sort order. For this reason, the Btree access method should be used when there is any locality of reference among keys. Locality of reference means that accessing one particular key in the Btree implies that the application is more likely to access keys near to the key being accessed, where "near" is defined by the sort order. For example, if keys are timestamps, and it is likely that a request for an 8AM timestamp will be followed by a request for a 9AM timestamp, the Btree access method is generally the right choice. Or, for example, if the keys are names, and the application will want to review all entries with the same last name, the Btree access method is again a good choice.
There is little difference in performance between the Hash and Btree access methods on small data sets, where all, or most of, the data set fits into the cache. However, when a data set is large enough that significant numbers of data pages no longer fit into the cache, then the Btree locality of reference described previously becomes important for performance reasons. For example, there is no locality of reference for the Hash access method, and so key "AAAAA" is as likely to be stored on the same database page with key "ZZZZZ" as with key "AAAAB". In the Btree access method, because items are sorted, key "AAAAA" is far more likely to be near key "AAAAB" than key "ZZZZZ". So, if the application exhibits locality of reference in its data requests, then the Btree page read into the cache to satisfy a request for key "AAAAA" is much more likely to be useful to satisfy subsequent requests from the application than the Hash page read into the cache to satisfy the same request. This means that for applications with locality of reference, the cache is generally much more effective for the Btree access method than the Hash access method, and the Btree access method will make many fewer I/O calls.
However, when a data set becomes even larger, the Hash access method can outperform the Btree access method. The reason for this is that Btrees contain more metadata pages than Hash databases. The data set can grow so large that metadata pages begin to dominate the cache for the Btree access method. If this happens, the Btree can be forced to do an I/O for each data request because the probability that any particular data page is already in the cache becomes quite small. Because the Hash access method has fewer metadata pages, its cache stays "hotter" longer in the presence of large data sets. In addition, once the data set is so large that both the Btree and Hash access methods are almost certainly doing an I/O for each random data request, the fact that Hash does not have to walk several internal pages as part of a key search becomes a performance advantage for the Hash access method as well.
Application data access patterns strongly affect all of these behaviors, for example, accessing the data by walking a cursor through the database will greatly mitigate the large data set behavior describe above because each I/O into the cache will satisfy a fairly large number of subsequent data requests.
In the absence of information on application data and data access patterns, for small data sets either the Btree or Hash access methods will suffice. For data sets larger than the cache, we normally recommend using the Btree access method. If you have truly large data, then the Hash access method may be a better choice. The db_stat utility is a useful tool for monitoring how well your cache is performing.
The Queue or Recno access methods should be used when logical record numbers are the primary key used for data access. The advantage of the Queue access method is that it performs record level locking and for this reason supports significantly higher levels of concurrency than the Recno access method. The advantage of the Recno access method is that it supports a number of additional features beyond those supported by the Queue access method, such as variable-length records and support for backing flat-text files.
Logical record numbers can be mutable or fixed: mutable, where logical record numbers can change as records are deleted or inserted, and fixed, where record numbers never change regardless of the database operation. It is possible to store and retrieve records based on logical record numbers in the Btree access method. However, those record numbers are always mutable, and as records are deleted or inserted, the logical record number for other records in the database will change. The Queue access method always runs in fixed mode, and logical record numbers never change regardless of the database operation. The Recno access method can be configured to run in either mutable or fixed mode.
In addition, the Recno access method provides support for databases whose permanent storage is a flat text file and the database is used as a fast, temporary storage area while the data is being read or modified.