The following topics describe the features of the MicroKernel Engine:

The Btrieve API is a low-level interface to the MicroKernel Engine that embodies functional aspects of the database design that might be otherwise transparent in higher-level interfaces such as SQL, Java, or ODBC. For example, the SQL interface operates independently of how the data is physically stored. However, the MicroKernel Engine developer must consider lower level aspects such as page size, physical and logical currency, type verification, and data validation. Despite these low-level considerations, the Btrieve API provides excellent flexibility and control over the data.

The MicroKernel Engine stores information in files, which can be up to terabytes in size for the 13.0 and 16.0 file versions, (256 GB for the 9.5 version, 128 GB for earlier 9.x versions, and 64 GB for other earlier versions). Inside each data file are records, which contain bytes of data. A file can contain quintillions of records.

The data in a record might represent an employee name, ID, address, phone number, rate of pay, and so on. However, the MicroKernel Engine interprets a record only as a collection of bytes; it does not recognize logically discrete pieces of information within a record. To the MicroKernel Engine, a last name, first name, employee ID, and so on do not exist inside a record.

The only discrete portions of information that the MicroKernel Engine recognizes in a record are keys. Keys provide both fast, direct access to records and a means of sorting records by key values. Because the MicroKernel Engine has no way of knowing the structure of the records in each file, you define each key by identifying the following:

You can create or drop keys at any time. For each key defined in a data file, the MicroKernel Engine builds an index. The index is stored inside the data file itself. The index maps each key value in the file to an offset in the actual data. Normally, when accessing or sorting data, the MicroKernel Engine does not search through all the records in the file. Instead, it searches the index and then manipulates only those records appropriate to the request.

You can create indexes when you create the data file, or any time thereafter. When you create a data file, you can define one or more keys for the MicroKernel Engine to use in building indexes.

You can also define external indexes after creating a file. An external index file is a standard data file that contains records sorted by the key you specify. Each record consists of the following:

Positioning rules (guidelines governing which record is current, which is next, and so on) are the same, regardless of when you create an index.

If you create an index at the same time that you create the file, the MicroKernel Engine stores duplicate key values in the chronological order in which the records are inserted into the file. If you create an index for a file that already exists, the MicroKernel Engine stores duplicate key values in the physical order of the corresponding records in the file at the time the index is created. How the MicroKernel Engine stores duplicate key values in an index also depends on whether the key is linked-duplicatable or repeating-duplicatable. For more information, see Duplicatability.

You can delete, or drop, an index when your application no longer needs it. The space that the index used in the file is freed for data or for other index pages. (However, this free space remains allocated to the file; you will not see a reduction in physical file size after dropping an index.)

See Designing a Database for specific information about defining keys.

Before an end user can run your MicroKernel Engine application, a version of MicroKernel Engine must be available to the end user's computer. You should provide the end user with information about any prerequisite the MicroKernel Engine software versions and configurations that your application requires.

End users may need to know the following information about your MicroKernel Engine application. Include this information in the documentation you provide with your MicroKernel Engine application.

For descriptions of configuration options, see Advanced Operations Guide.

The following topics cover pages and how the MicroKernel Engine handles them:

Files consist of a series of pages. A page is the unit of storage that the database transfers between memory and disk. A file is composed of the following types of pages:

All 6.0 and later files have FCR and PAT pages. Standard files also contain data and index pages, and optionally, variable and ACS pages. Data-Only Files contain no index pages. Key-Only Files contain no data pages.

You specify a fixed page size when you create a file. The page size you can specify, the file overhead, and so forth, depends on a variety of factors, including the file format. See Designing a Database for information on page sizes. The following sections provide an overview:

The page size you specify should satisfy the following criteria:

To make the most efficient use of modern operating systems, you should choose a larger page size. The smaller page sizes were used when DOS was the prominent operating system (when a sector was 512 bytes and all I/O occurred in multiples of 512). This is no longer the case. Both 32-bit and 64-bit operating systems move data around their cache in blocks of 4096 bytes or larger. CD ROM drives are read in blocks of 2048 bytes.

The MicroKernel Engine indexes are most efficient when a page size of 4096 bytes or larger is used. The key will have more branches per node and thus will require fewer reads to find the correct record address. This is important if the application is doing random reads using a key. This is not important when an application accesses the file in a sequential manner either by key or by record.

A good reason for having smaller page sizes is to avoid contention. With fewer records in each page it becomes less likely that different engines or transactions will need the same page at the same time. If a file has relatively few records, and the records are small, you may want to choose a small page size. The larger the file, the less likely contention will happen.

Another potential problem with large page sizes is specific to version 7.0 and later files. There is a maximum of 256 records or variable-length sections that can fit on the same data page. If you have short or compressed records, or short variable-length sections, you can easily reach the limit while you still have hundreds of bytes available on every page. The result is a much larger file than needed. By knowing your record size, you can calculate how big of an issue this is.

The database designer must choose between these conflicting needs. A reference table that is not changed very often, but is searched or scanned most of the time, should have larger page sizes. A transaction file which is inserted and updated within transactions should have smaller page sizes.

Only careful consideration of all factors can give the right answer to what the page size should be. For more information about choosing a page size, see Choosing a Page Size.

The Btrieve API supports three data file types, large files to a maximum file size of terabytes for the 13.0 and 16.0 file versions, 256 GB for 9.5 files, 128 GB for earlier 9.x versions, and 64 GB for other earlier versions, as well as long file names. The following topics cover these features:

A standard 7.x or later data file contains two FCR pages followed by a number of PAT pages, index pages, data pages, and possibly variable and ACS pages. You can create a standard file for use with either fixed- or variable-length records. Because standard files contain all the index structures and data records, The MicroKernel Engine can dynamically maintain all the index information for the records in the file.

When you create a data-only file, you do not specify any key information, and Zen does not allocate index pages for the file. This results in a smaller initial file size than for standard files. You can add keys to a data-only file after creating the file.

Key-only files contain only FCR pages followed by a number of PAT pages and index pages. In addition, if you have defined referential integrity constraints on the file, the file may contain one or more variable pages.

Key-only files include only one key, and the entire record is stored with the key, so no data pages are required. Key-only files are useful when your records contain a single key and that key takes up most of each record. Another common use for a key-only file is as an external index for a standard data file.

The following restrictions apply to key-only files:

The MicroKernel Engine supports file sizes up to terabytes for the 13.0 and 16.0 file versions, 256 GB for 9.5 files, 128 GB for earlier 9.x versions, and 64 GB for other earlier versions. However, many operating systems do not support single files this large. In order to support files larger than the operating system file size limit, the MicroKernel Engine breaks up large files into smaller files that the operating system can support. A large, logical file is called an extended file. The smaller, physical files that comprise an extended file are called extension files. The base file is an original data file that has become too large to support as a single, physical file. Nonextended (that is, nonsegmented) files provide more efficient I/O and, therefore, increased performance.

You can choose to not to automatically extend 9.x format files or later at 2 GB. To change the segment operation setting access the configuration settings in the Zen Control Center (ZenCC) as described in Configuration Using ZenCC in Advanced Operations Guide. From there you can set the Limit Segment Size to 2 GB option. This setting as no effect on 13.0 and 16.0 files, which are never segmented.

If this option is unselected, PSQL 9.x files will not be segmented automatically at 2 GB. Data files from version 8.x and earlier will continue to be extended when they reach 2 GB. If your files are already extended, they will remain segmented.

Regardless of the configuration setting, all files will continue to be extended based on the file size limitations of the current operating system.

For information about backing up files, including extended files, see Backing Up Your Files.

The MicroKernel Engine supports long file names whose length is less than or equal to 255 bytes. The following items must conform to this upper limit:

The file name can contain spaces unless the Embedded Spaces client configuration option is disabled. The default setting is On. See Advanced Operations Guide (Long File Names and Embedded Spaces Support).

When the MicroKernel Engine generates new files based on an existing file name, such as with Large Files or during Archival Logging or Continuous Operations (for more information, see Logging, Backup, and Restore in Advanced Operations Guide), the new file name includes as much of the original file name as possible and an appropriate file extension, as in the following examples:

When using 7.x or a later file format, you can use the following data types when you define a key:

If you are using the 6.x file format, you can use all the preceding types to define a key except for CURRENCY, TIMESTAMP, and WZSTRING.

If you are using a file format before 6.x, NUMERICSA and NUMERICSTS are not available as data or key types.

For more information, see Data Types in SQL Engine Reference.

The following sections describe the attributes you can assign when you define a key:

These topics contains information on attributes that you can assign to keys:

Keys can consist of one or more segments in each record. A segment can be any set of contiguous bytes in the record. The key type and sort order can be different for each segment in the key.

The number of index segments that you may use depends on the file's page size.

See status codes 26: The number of keys specified is invalid and 29: The key length is invalid for related information about index segments and the MicroKernel Engine.

The total length of a key is the sum of the length of the key segments, and the maximum length is 255 bytes for 13.0 format files and earlier, or 1024 bytes for 16.0 format files. Different key segments can overlap each other in the record.

When a segmented key is a nonduplicatable key, the combination of the segments must form a unique value; however, individual segments may contain duplicates. When you are defining this type of segmented key, each segment has duplicates=no as a key-level attribute even though that particular segment may have duplicates. To ensure that a particular segment is always unique, define it as a separate nonduplicatable key in addition to the segmented key definition.

When issuing a call to the MicroKernel Engine, the format of the key buffer must be able to accommodate the key specified by the key number. So, if defined keynumber=0 and key 0 is a 4-byte integer, the key buffer parameter can be any of the following:

Basically, the MicroKernel Engine gets a pointer to a memory location to be used as a key buffer. The MicroKernel Engine expects that memory location to have a data value corresponding to the specified key number for certain operations, such as Get Equal. In addition, the MicroKernel Engine may write data out to that location, and the data written will be a key value corresponding to the specified key number. In this situation, the memory location must be allocated large enough to accommodate the entire key value.

To the MicroKernel Engine, a key is a single collection of data, even if it is made up of multiple segments. The segment feature allows you to combine noncontiguous bytes of data together as a single key. It also allows you to apply different sorting rules (as dictated by the supported data types) to different portions of the key data. The data type associated with a key segment is used typically as a sorting rule – it tells the MicroKernel Engine how to compare two values to determine which one is larger. Data types are not used to validate data.

The MicroKernel Engine always deals with an entire key, not a key segment. To work with any key, set up a key buffer that is large enough to hold the entire key. Some applications define a generic 255 byte buffer to use on all calls to the MicroKernel Engine; this is the maximum size of a key in 13.0 format files and earlier, and is often sufficient. When data is returned in this key buffer, the application usually copies data out of the generic buffer into an application variable or structure declared as the same type(s) as the key segment(s). Alternatively, pass a key buffer parameter (simple variable or structure variable) that directly corresponds to the key.

For example, suppose you want to read a record and only know the value of the first segment of the key, but not all segments. You can still utilize that key to find the data. However, you still have to pass in an entire key buffer corresponding to all segments. Because you only know part of the key value, you cannot use the Get Equal call. You have to use the Get Greater Or Equal call. In this case, initialize the key buffer with as many key values as you know and then specify low or null values for the unknown key segments.

For example, given a key 1 definition of three segments corresponding to data values ulElement2, ulElement3, and ulElement5, if you know what value you want for ulElement2, you would initialize your key buffer as:

and then pass &SampleKey1 as the key buffer parameter on a Get Greater Or Equal call. When the MicroKernel Engine completes the call and a record is found, status code 0 is returned, the corresponding data record is returned, and the key buffer is set to have the key value including all three segments.

Zen supports two methods for handling duplicate key values: linked (the default) and repeating. With linked-duplicatable keys, the MicroKernel Engine uses a pair of pointers on the index page to identify the chronologically first and last records with the same key value. Additionally, the MicroKernel Engine uses a pair of pointers in each record on the data page to identify the chronologically previous and next records with the same key value. The key value is stored once, and only on the index page.

With repeating-duplicatable keys, the MicroKernel Engine uses a single pointer on the index page to identify the corresponding record on the data page. The key value is stored on both the index page and the data page. For more information on duplicate keys, see Duplicatable Keys.

If you define a key as modifiable, the MicroKernel Engine enables you to change the value of a key even after the record is inserted. If one segment of a key is modifiable, all the segments must be modifiable.

By default, the MicroKernel Engine sorts key values in ascending order (lowest to highest). However, you can specify that the MicroKernel Engine order the key values in descending order (highest to lowest).

When you perform a Get Greater (8) operation on a descending key, the MicroKernel Engine returns the record corresponding to the first key value that is lower than the key value you specify in the key buffer. For example, consider a file that has 10 records and a descending key of type INTEGER. The actual values stored in the 10 records for the descending key are the integers 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9. If the current record's key value is 5 and you perform a Get Greater operation, the MicroKernel Engine returns the record containing the key value 4.

Similarly, when you perform a Get Less Than (10) operation using a descending key, the MicroKernel Engine returns the record with the next higher value than the one you specify in the key buffer. Using the preceding example, if the current record's descending key has a value of 5 and you perform a Get Less Than operation, the MicroKernel Engine returns the record containing the key value 6.

By default, the MicroKernel Engine is case sensitive when sorting string keys; that is, it sorts uppercase letters before lowercase letters. When you define a key to be case insensitive, the MicroKernel Engine sorts values without distinguishing case. Case sensitivity does not apply if the key has an alternate collating sequence (ACS).

Zen contains two ways of identifying a column of data as a Null value. The original type of null value, called a legacy null, has been used in the MicroKernel Engine for years. The newer type of null is the true null. This topic describes the legacy null and then details use of true nulls in the MicroKernel Engine.

The original method of defining a nullable field is referred to as pseudo-null or legacy-null. It is based on the premise that if the entire field is full of a particular byte value, typically ASCII zero, then the field is considered Null. The byte value is defined in the key definition supplied when creating the index. Using the MicroKernel Engine, the only thing that the MicroKernel Engine can do with this knowledge is to decide whether or not to include the field in an index. There are no special sorting rules for Legacy Nulls since they are just values that sort just like all the other values, despite their special meaning.

If a key description contains the flag for All Segments Null (0x0008), then the key value is not put into the index if each and every segment in the key is considered to be null by having the null byte in every byte of the field. Likewise, if a key description contains the flag for Any Segment Null (0x0200), then the key value is not put into the index if one or more key segment is considered to be Null by the same rule

The Relational Engine never uses these flags in the indices that it defines, since in order to make joins between tables, all records in the table must be accessible through the index.

Starting with Pervasive.SQL 2000, a new type of Null Indicator was introduced called true null.

True nulls are implemented in the MicroKernel Engine by preceding a nullable field with a one byte Null Indicator Segment (NIS). This is an extra byte of data outside the normal column width that indicates whether the column is Null or not. A value of zero in this byte indicates that the column associated with it is normal, or Not Null. Any other value in this byte indicates that the column value is Null.

With true nulls, unlike legacy nulls, you can tell the difference between an integer that is zero and one that is Null. This is true for any type of number field. You can even distinguish a string field that has a zero length string in it from one that should be identified as null, if there is a need for such a distinction.

The Relational Engine can identify and use true null columns whether or not there is an index defined on them, but a basic data file only identifies the fields that are included in keys.

You can define true null fields within MicroKernel Engine keys by adding a Null Indicator Segment (NIS) before the nullable field in the key definition of a Create (14) or Create Index (31) operation. See Rules for True Null Keys for the rules regarding true null keys.

The MicroKernel Engine does not enforce any restrictions on the offset of the NIS, whereas the Relational Engine assumes that it immediately precedes the nullable field. As such, it is recommended to structure the fields within your record to make room for the NIS in the byte preceding any field that will use an NIS. This preserves your ability to access these tables through SQL should you need to do so.

The following rules must be followed when using this new key type:

Any nonzero value is considered an indicator that the following segment is null. By default, the MicroKernel Engine makes no distinction between nonzero numbers. The Zen relational engine currently uses only a value of 1 in this field to indicate a null. It is possible, however, to make a distinction between different types of nulls. This can be done by using the Case Insensitive flag on the NIS. Since this key flag is normally only applicable to the various string and character fields, it is overloaded to have the special meaning of DISTINCT when used with an NIS. It means that different NIS values should be treated distinctly and sorted separately. Actian Corporation reserves the use of the first 15 values for future use. If you want to apply a special meaning to various types of nulls in your application, please use NIS values greater than 16. For example, more specific null definitions could be:

When you add the DISTINCT flag (Case Insensitive) to the NIS, these nonzero values will be sorted separately as distinct values.

A true null field has a nondeterminate value. In other words, its value cannot be known. According to this definition, no two Null values are equal to each other, nor are they equal to any other key value. Yet the MicroKernel Engine must group these Null values together and you must be able to find key values that are equal to Null. To accomplish this, the MicroKernel Engine interprets true null values as if they are equal to each other, depending on the purpose of the comparison. When sorting, and finding a place for the Null values in an index, true null values are grouped together as if they were equal to each other. But when trying to determine if a value already exists in a unique index, true nulls are not equal to each other.

Any nonzero value in the NIS means the following field is Null. The default behavior is to treat all nonzero values in the NIS as if they were the same value and interpret them to indicate that the nullable field is Null. As such, if you insert records that contain a variety of nonzero values in the NIS and a variety of values in the nullable field that follows, they will all be interpreted as the same value, and will be sorted as a collection of duplicates.

This section discusses the results of inserting several Null values into a Linked Duplicate key into a Linked Duplicate key.

Linked Duplicates contains a single key entry for each unique value, with two record address pointers; one for the first duplicate record and one for the last record in the duplicate chain. Each record contains 8 bytes of overhead consisting of pointers to the previous and next records in the chain. Each new duplicate value is added at the end of the chain, thus ensuring that the duplicate records are linked in the order they were inserted. All true null values are considered duplicates for the purpose of adding them to an index, so they all will be linked to the same chain in the order they were inserted. Even if each record contained different byte values in the NIS and the associated nullable field, there will only be one key entry pointing to the first and last record in this chain. If the NIS key segment is defined as descending, this key entry will occur first in the index. Otherwise, it will occur last.

Repeating Duplicate Keys contain an actual key entry for each record represented in the index. There is no overhead in the record itself and for each record, there is a key entry that points to it. Duplicate values in this kind of index are sorted by the physical record address to which they point. This means that the order of duplicates is unpredictable, especially in a highly concurrent environment where random records are being inserted and deleted by many clients.

True Null values are interpreted as if they are duplicates and are sorted not by the bytes found in the nullable field, but rather by the record address. So when using repeating duplicate keys, the records containing true null values are grouped together, but in a random fashion. If the NIS segment is descending, they will occur first in the index, otherwise, they will occur last.

In the MicroKernel Engine, if you define an index without either duplicate flag, the index must contain only unique values. But since the value of a true null field is indeterminate, they should not be considered duplicates. For this reason, the MicroKernel Engine allows multiple true null values to be entered into a unique key, assuming that once the value is assigned with an update operation, then the uniqueness of the key can be determined. But for the purposes of sorting these values in the index, the MicroKernel Engine groups them all together as if they were duplicates. So the section of the index containing the true null values resembles a Repeating Duplicate index. The nulls are sorted together according to the physical record address, the order of which cannot be predicted.

Once you put a value into a nonmodifiable key, it cannot be changed. But because a true null value does not have an actual value, the MicroKernel Engine allows you to insert a record with a true null value in any or all fields defined in true null indexes, and then later change those field values in an update operation from null to nonnull. But once any field has become nonnull, the nonmodifiability is enforced and it cannot be changed again, even if to establish the field as null again.

Even though true null values are indeterminate and are not considered equal to each other, it is possible to locate a record with a true null key segment.

The various Get operations can address true null keys by using this sequence:

The following list shows the expected behavior from the Get operations:

This is consistent with the behavior of the Get operations for normal duplicate values.

It is possible to distinguish between different values in the NIS byte. The default behavior, as indicated, is that all nonzero values in the NIS are considered to be the same thing, and whatever the NIS contains, if it is not zero, the nullable field is Null. The Relational Engine currently uses this default behavior on all true null index segments that it creates.

However, if you want to store different kinds of Null values in your table, then you can add the NOCASE flag (0x0400) to the key definition of the NIS segment. Hereafter, we will call this the DISTINCT flag. When you do this, the MicroKernel Engine will treat different NIS values as different or distinct from each other.

Distinct True Null segments are sorted in groups by their NIS value. The same rules apply as discussed above when building the various types of indexes. A linked duplicate key will have a single entry for each distinct NIS value with a pointer to the first and last occurrence of that type of Null. Repeating Duplicates and Unique keys will also group the null records by their distinct NIS value. Descending Keys have the highest NIS values grouped first, sorted down to the zero, or nonnull values. Ascending keys sort the nonnull records first, followed by NIS values of 1, then 2, and so on. Get operations pay attention to the value of the NIS. If you do a GetEQ using a key buffer where the NIS is 20, and all the NIS values in a Distinct True Null index are 1, then the MicroKernel Engine will not find any matching values.

Although the Relational Engine nor any Zen access method currently uses the DISTINCT flag when creating true null indexes, they might in the future. For this reason, NIS values 2 through 16 are reserved for future use, in case Zen needs to assign specific meanings to these 'types' of nulls. So if you use distinct null values for records accessed through the transactional Btrieve API, use values greater than 16.

Consider a multi-segmented True Null index containing two nullable string columns. The key would actually be defined as a four segment index. The first segment is an NIS, followed by the first nullable field, then the second NIS followed by the second nullable field. Now consider what would happen if the following records were put into the file.

plus a couple more of these records; "BBB", NULL

The Relational Engine always creates True Null index segments such that the NULL values will occur first. It does this by adding the Descending flag (0x0040) to each NIS segment. Let's assume that the descending flag is used on each NIS and on the second nullable field, but not the first nullable field. If so, these records would be sorted like this.

The nulls always occur before the nonnulls since both NIS are descending. But when the NIS is zero, i.e, the fields are nonnull, the first field is sorted ascending and the second is sorted descending.

The following is what would be returned by various Get operations;

GetLT "BBB", NULL returns record 8 "AAA", "AAA"

GetLE "BBB", NULL returns record 11 "BBB", NULL

GetEQ "BBB", NULL returns record 9 "BBB", NULL

GetGE "BBB", NULL returns record 9 "BBB", NULL

GetGT "BBB", NULL returns record 12 "BBB", "CCC "

The GetLE has the implication that you are looking to traverse the file in the reverse direction, so it returns the first occurrence of a key value that "matches" in the reverse direction. GetEQ and GetGE imply that you are moving in the forward direction.

As with legacy nulls, you can also apply the flag for "All Segments Null" (0x0008) or "Any Segment Null" (0x0200) to each segment of any index containing an NIS. When you insert a record, the MicroKernel Engine will determine of the nullable field is Null using the NIS. The same rules apply to determine if the key entry will be put into the index or not.

So you should not use these flags if you think that you might at some point want to access these files from SQL, where a goal might be to find any records "where column IS NULL". The Relational Engine will use the index to find the null records, but they will not be accessible through the index.

Extended operations allow your application to access fields in a table even if they do not have indexes created for them. You can apply a filter to the fields in your record, defining fields on the fly, using knowledge of the record from any source. Thus it is possible to define True Null fields in an extended operation and have the MicroKernel Engine apply the same comparison rules that it would when sorting these fields into an index.

You must define the extended operations filter just like you would define a key. Include a filter segment for the NIS followed by the nullable field. You must include the nullable field in the filter even if you are searching for a null value, where the content of the nullable field does not matter. The MicroKernel Engineneeds both filter segments so that a GetNextExtended can be optimized against an index path and it enforces this with status 62, indicating that a filter expression for an NIS was not followed by a non-NIS.

The only comparison operator you can use for an NIS is EQ or NE. You will get status 62 if you try to use any of the other comparison operators GT, GE, LT, and LE.

A status 62 occurring from a badly formed extended operation descriptor adds a system error Zen Event Log. These system errors are listed under True Nulls and Extended Operations to help you identify the reason for the status 62.

If you want to treat different NIS values distinctly, then add 128 to the comparison operator on the NIS field. This is the same bias value that you would use to indicate case insensitivity. And just like when defining an index, the case insensitive flag has been overloaded for Null Indicator key types to indicate that the nonzero values should be compared distinctly meaning that they should be distinguished from one another instead of treating them all the same.

If you are using the extended operations to get the best performance possible, you will be trying to search along a key path for specific limited ranges of key values. First, establish currency at the beginning of the range by using GetGE. Then follow that with GetNextExtended. Or, you can do a GetLE followed by GetPrevExtended. These extended operations can stop searching automatically when there is no more chance of finding any more values that match the filter. This is called extended operation optimization. If your filter can use optimization, it will be much more efficient because there may be a huge number of records that can be skipped and not read from the file. In order to create an optimized search, you need to be traversing the index in a direction where a limit exists. Also, your filter must exactly match the index, using AND instead of OR as segment connectors.

If you do a GetNextExtended on an ascending index, then an optimized filter can stop at the limit when the conditional operator is EQ, LT or LE. A search will have to look to the end of the file for values greater than a particular value going forward along an ascending index. Likewise, if it is a descending index, then it can stop at a limit when the conditional operator is EQ, GT or GE. This can get much more complicated when there are multiple fields in the search criteria. The simple way to think about it is that in order to optimize a filter, only the last segment can have any other conditional operator than EQ. This includes the NIS. If the conditional operator on an NIS is NE, the filter can only be optimized up to the previous filter segment.

Exactly matching the index means that each filter expression should follow the order of the segments in the index and have the same offset, length, key type, case sensitivity (or distinct flag), and ACS specification. Otherwise, extended operations cannot be optimized.

True nulls are implemented in the Relational Engine through the use of the Null Indicator key type and follow the rules described above. The MicroKernel Engine applications can also use this key type to identify the nullness of a nullable field regardless of its contents. This provides a way to identify null integers and other number data types, and fully manage these nullable fields.

Status 62 occurring on an extended operation indicates that the descriptor is incorrect. Sometimes, it may be difficult to determine what exactly is wrong with the descriptor. The database engine adds a line in the Zen event log that can be used to determine the exact problem. The event log entry will look similar to the following:

The following numbers immediately after the system error 301 through 318 identify the problem.

You can use an alternate collating sequence (ACS) to sort string keys (types STRING, LSTRING, and ZSTRING) differently from the standard ASCII collating sequence. By using one or more ACS files, you can sort keys as follows:

Files can have a different ACS for each key in the file, but only one ACS per key. Therefore, if the key is segmented, each segment must use either the ACS specified for that key or no ACS at all. For a file in which a key has an ACS designated for some segments but not for others, the MicroKernel Engine sorts only the segments that specify the ACS.

To create an ACS to sort string values differently from the ASCII standard, use the format shown in the following table.

Because ACS files are created using a hex editor or in a MicroKernel Engine application, they are useful mainly to application developers and not usually created by end users.

Here is an example of a 9-byte header and a 256-byte body that represent a collating sequence named UPPER. The header appears as follows:

The 256-byte body looks like this, with the exception of the offset values in the leftmost column:

The header and body forming this ACS are shipped with Zen as the file UPPER.ALT. UPPER.ALT provides a way to sort keys without regard to case. (You can define a key to be case-insensitive; even so, UPPER provides a good example when writing your own ACS.)

Offsets 0x61 through 0x7A in the example have been altered from the standard ASCII collating sequence. For standard ASCII, offset 0x61 contains a value of 0x61 (lowercase a). When a key is sorted with the UPPER ACS, the MicroKernel Engine sorts lowercase a (0x61) with the collation weight at offset 0x61: 0x41. Thus, the lowercase a is sorted as if it were uppercase A (0x41). Therefore, for sorting purposes UPPER converts all lowercase letters to their uppercase equivalents when sorting a key.

The following 256-byte body performs the same function as the UPPER.ALT body except that ASCII characters preceding the ASCII space (0x20) are now sorted after all other ASCII characters:

In this body, different collating weights have been assigned so that a character's weight no longer equals its ASCII value. For example, offset 0x20, representing the ASCII space character, has a collating weight of 0x00; offset 0x41, representing the ASCII uppercase A, has a collating weight of 0x21.

To sort keys without regard to case, offsets 0x61 through 0x7A in the last example have been altered. As in the body for UPPER.ALT, offset 0x61 has the same collating weight as offset 0x41: 0x21. By having the same collating weight, offset 0x41 (A) sorts the same as offset 0x61 (a).

To specify an ACS that sorts string values using an ISO-defined, language-specific collating sequence, you must specify an ISR table name, as shown in the following examples:

The ISR tables are installed with Zen and based on ISO-standard locale tables. ISR tables are stored in the collate.cfg file, which is installed with the Zen database engine. Data files can share a single ISR.

For examples, see Sample Collations Using International Sorting Rules.

When you create an index using either Create (14) or Create Index (31), you must provide the key specification structure.

The following table shows the data buffer structure of the key specification. Each key specification block is 16 bytes for BTRV type entry points or 24 bytes for BTRVEX types. Where two data types are given, but first is used with BTRV and the second with BTRVEX.

The following table gives values for key flags in the key specification block.

The following segment-specific key flags should be on, with the NIS: SEGMENTED (0x0010), EXTENDED DATA TYPE (0x0100).

The following flags should be off: None.

The following flags are ignored: BINARY (0x0004), ACS (0x0020).

True null support comes with a few limitations:

A key concept in using the Btrieve Login API or the implicit login functionality via the Create or Open functions is the database Uniform Resource Identifier (URI). The URI provides a syntax to describe the address of a database resource on a server.

This topic describes the syntax and semantics of the URIs used in Btrieve APIs.

URIs use the following syntax:

The special characters separating the elements are required, even if an element is omitted. The following table lists the elements of this URI.

The database engine enforces a precedence level on the parameters file, table, and dbfile when more than one of them is specified in a URI. That is, after parsing, the database engine leaves the parameter with the highest precedence. If two or more have the same precedence, the last parameter in the URI is returned after parsing.

The order of precedence from highest to lowest is file, table, and dbfile.

As with any URI, certain nonalphanumeric characters have special significance in the URI syntax. If you wish to use one of these characters within one of the elements in the URI, you must use an escape sequence to identify the character as actual text rather than a special character. An escape sequence is simply another special character or character combination that represents the plain text equivalent of a special character.

The table below shows the special characters supported by the MicroKernel Engine URI syntax, and their associated escape sequences (represented by the percent sign and the hexadecimal value for the specified character).

Although the space character is reserved in the URI specification, it can be used without quotes and without escape sequencing because it is not used as a delimiter. The other symbols in the table above are used as delimiters and therefore must be escaped.

This section shows examples of URIs using escape sequences to identify special characters used within the field values.

Note that an empty user name or password is different than no user name or password. For example, btrv://@host/ has an empty user name, while btrv://host/ has no user name, and btrv://sam@host/?pwd= has a user name sam with an empty password.

Some URIs allow the use of user:password syntax. However, the password is then transmitted as clear text. To prevent the transmission of the password as clear text, the Zen database URI ignores the password if one is provided using the user:password syntax. Use the pwd= parameter to provide a password, which the Zen client changes into an encrypted password before transmission.

Some URIs allow for server based naming authority with a syntax of user@host:port. The Zen database URI does support specifying a port element.

A URL (Uniform Resource Locator) is simply the address of a file or resource on the Internet. The database URI uses the same notion to address a database on a server. This topic gives examples of the syntax and semantics of URIs for Zen databases, particularly using the MicroKernel Engine access.

The URI and UNC syntax does not allow certain special characters, such as colons. Since raw IPv6 addresses use colons, different methods of handling UNC paths and URI connections are available. Zen supports IPv6-literal.net Names and Bracketed IPv6 Addresses.

See Drive-based Formats in Getting Started with Zen.

Zen accepts Shift-JIS (Japanese Industrial Standard) encoded double-byte characters in file paths. (Shift-JIS is an encoding technique commonly used for Japanese computers.) In addition, you can store Shift-JIS double-byte characters in records and sort them using the Japanese ISR table described in International Sort Rules. Other multi-byte characters can be stored in records, although ISR tables are currently not available to sort these records according to culturally correct rules. Your use of double-byte characters does not affect the operation of Zen applications.

All records contain the record length, a fixed-length portion which must be large enough to contain all the data (including keys) for a record, plus the overhead required to store a record on a data page.

See Optimum Page Size For Minimizing Disk Space and Record Overhead in Bytes With Record Compression for how many bytes of overhead you must add to the logical record length to obtain a physical record length.

The following table lists the maximum record size for fixed-length records.

Note that the database engine turns on data compression for the file if the file uses system data and the record length exceeds the limit shown in the table above.

Optionally, the records in a file can contain a variable-length portion. A variable-length record has a fixed-length portion that is the same size in every record and a variable-length portion that can be a different size in each record. When you create a file that uses variablelength records, the fixed-length amount is the minimum length of each record; you do not define the maximum record length.

Theoretically, the maximum length of variable-length records is limited only by the MicroKernel Engine file size limit: up to terabytes for 13.0 and 16.0 version files, 256 GB for 9.5 files, 128 GB for earlier 9.x versions, and 64 GB for other earlier versions. In practice, the maximum is limited by such factors as the operating system, system resources, and the record access method. If you retrieve, update, or insert whole records, then the data buffer length parameter, a 16-bit unsigned integer, limits the record length to 65535.

A data buffer is a MicroKernel Engine function parameter that you use to transfer various information depending on the operation being performed. A data buffer can contain all or part of a record, a file specification, and so forth. See Designing a Database for more information on data buffers.

If your file uses very large records, consider using variable-tail allocation tables (VATs) in the file. A VAT, which is implemented as a linked list, is an array of pointers to the variable-length portion of the record. VATs accelerate random access to portions of very large records. Some examples of very large records are binary large objects (BLOBs) and graphics.

For files that contain very large variable-length records, the MicroKernel Engine splits the record over many variable pages and connects the pages using a linked list called a variable tail. If an application uses chunk operations to access a part of a record and that part of the record begins at an offset well beyond the beginning of the record itself, the MicroKernel Engine may spend considerable time reading the variable-tail linked list to seek that offset. To limit such seek time, you can specify that the file use VATs. the MicroKernel Engine stores the VAT on variable pages. In a file containing a VAT, each record that has a variable-length portion has its own VAT.

The MicroKernel Engine uses VATs not only to accelerate random access, but also to limit the size of the compression buffer used during data compression. If your files use data compression, you may want to use VATs in the file.

The following features support concurrent access while ensuring the integrity of your files in a multi-user environment:

Applications can explicitly lock either one record at a time (single record lock) or multiple records at once (multiple record lock). When an application specifies a record lock, the application can also apply a wait or no-wait condition. When an application requests a no-wait lock on a record that is currently unavailable (either the record is already locked by another application or the whole file is locked by an exclusive transaction), the MicroKernel Engine does not grant the lock.

When an application requests a wait lock on a record that is unavailable, the MicroKernel Engine checks for a deadlock condition. You can configure the MicroKernel Engine to wait before returning a deadlock detection status code. Doing so improves performance in multi-user situations by allowing the MicroKernel Engine to wait internally, rather than forcing the application to retry the operation.

If you have a number of modifications to make to a file and you must be sure that either all or none of those modifications are made, include the operations for making those modifications in a transaction. By defining explicit transactions, you can force the MicroKernel Engine to treat multiple operations as an atomic unit. Other users cannot see the changes made to a file until the transaction ends. The MicroKernel Engine supports two types of transactions: exclusive and concurrent.

In an exclusive transaction, the MicroKernel Engine locks the entire data file when you insert, update, or delete a record in that file. Other applications (or other instances of the same application) can open the file and read its records, but they cannot modify the file. The file remains locked until the application ends or aborts the transaction.

In a concurrent transaction, the MicroKernel Engine can lock either records or pages in the file, depending on the operation you are performing. The MicroKernel Engine enables multiple applications (or multiple instances of the same application) to perform modifications inside concurrent transactions in different parts of the same file simultaneously, as long as those modifications do not affect other previously locked portions of the file. The record or page remains locked until the application ends or aborts the transaction. Concurrent transactions are available only for 6.0 and later files.

Clients can still read records from a file even if a concurrent transaction has locked the requested record. However, these clients cannot be operating from within an exclusive transaction. Also, they cannot apply a lock bias to their read operation if the file containing the requested record is currently locked by an exclusive transaction, or if a concurrent transaction has locked the requested record.

When a client reads a record using an exclusive lock, the MicroKernel Engine locks only the individual record; the rest of the page on which the record resides remains unlocked.

When you use exclusive transactions, the MicroKernel Engine causes other clients to implicitly wait on the locked file unless the No Wait bias is added to the Begin Transaction (19 or 1019) operation. The application seems to hang during this implicit wait state. If these exclusive transactions are short lived, you may not notice the wait time. However, the overall effect of many clients involved in implicit waits results in using a large amount of CPU time. Additionally, multiple position blocks in the same file share locks.

Exclusive transactions involved in implicit waits also waste network bandwidth. The MicroKernel Engine waits about one second before returning to the requester. The requester recognizes a wait condition and returns the operation to the MicroKernel Engine. Thus, exclusive transactions also can cause extra network traffic.

The amount of extra CPU cycles and network traffic increase exponentially with the number of clients waiting on the locked file combined with the length of time involved in the exclusive transaction.

You can configure the MicroKernel Engine to guarantee Transaction Durability and atomicity by logging all operations to a single transaction log. Transaction durability is the assurance that the MicroKernel Engine finishes writing to the log when a client issues the End Transaction operation and before the MicroKernel Engine returns a successful status code to the client. Atomicity ensures that if a given statement does not execute to completion, then the statement does not leave partial or ambiguous effects in the database, thereby ensuring the integrity of your database by keeping it in a stable state.

If you want atomicity without the overhead of transaction durability, you can use transaction logging feature in PSQL V8 and later releases. See Advanced Operations Guide for more information on Transaction Logging.

In a default installation, the transaction log is in C:\ProgramData\Actian\Zen\logs. The log must exist on the same machine as the Zen engine. You can change the location using the transaction log directory configuration option in ZenCC by right-clicking the MicroKernel Engine and selecting Properties > Directories.

The MicroKernel Engine maintains the transaction log in one or more physical files, called log segments. The MicroKernel Engine starts a new log segment when the current log segment reaches a user-defined size limit, no files have pending changes, and the MicroKernel Engine has finished a system transaction.

All transaction log segments have a .log extension. The MicroKernel Engine names log segments with consecutive, 8-character hexadecimal names, such as 00000001.log, 00000002.log, and so on.

To improve performance on specific files, you can open a file in Accelerated mode. (Version 6.x MicroKernel Engine accepted Accelerated open requests, but interpreted them as Normal open requests.) When you open a file in Accelerated mode, the MicroKernel Engine does not perform transaction logging on the file.

Zen uses a 7.x transaction log file format. In order for the MicroKernel Engine to log transactions on a file, the file must contain a log key, which is a unique (non-duplicatable) key that the MicroKernel Engine can use to track the record in the log. For files that have at least one unique (non-duplicatable) key, the MicroKernel Engine uses one of the unique keys already defined in the file.

For files that do not have any unique keys, the MicroKernel Engine can include system data upon file creation. The MicroKernel Engine includes system data in a file only if the file uses the 7.x file format or later and if at the time the file is created, the MicroKernel Engine is configured to include system data in files upon creation. System data is defined as an 8-byte binary value with the key number 125. Even if a file has a unique, user-defined key, you may want to use system data (also called the system-defined log key), to protect against a user dropping an index.

The database engine turns on data compression for the file if the file uses system data and the record length exceeds the limit described under Record Length.

The MicroKernel Engine adds system data only at file creation. If you have existing files for which you want to add system data, than after you turn on the System Data configuration option, use the Rebuild tool as described in Converting Data Files.

The MicroKernel Engine uses shadow paging to protect 6.0 and later files from corruption in case of a system failure. When a client needs to change a page (either inside or outside a transaction), the MicroKernel Engine selects a free, unused physical location in the data file itself and writes a new page image, called a shadow page, to this new location. During a single MicroKernel Engine operation, the MicroKernel Engine might create several shadow pages all the same size as the original logical pages.

When the changes are committed (either when the operation is complete or the transaction ends), the MicroKernel Engine makes the shadow pages current, and the original pages become available for reuse. The MicroKernel Engine stores a map, which is the Page Allocation Table, in the file to keep track of the valid and reusable pages. If a system failure occurs before the changes are committed, the MicroKernel Engine does not update the PAT, thereby dropping the shadow pages and reverting to using the current pages, which are still in their original condition.

When a client operates inside a transaction, the shadow pages corresponding to the original logical pages are visible only to that client. If other clients need to access the same logical pages, they see the original (unchanged) pages – that is, they do not see the first client's uncommitted changes. Shadow paging thus enhances reliability because the original file is always valid and internally consistent.

Backing up your files regularly is an important step in protecting your data.

For information on backing up your files, please see the following topic in Advanced Operations Guide: Logging, Backup, and Restore.

Event logging is a feature in Zen that uses a log file to store informational, warning, and error messages from the MicroKernel Engine, SQL interface, and utility components.

See Reviewing Message Logs in Advanced Operations Guide for details.

The MicroKernel Engine provides the following features for enhancing performance:

To gain better performance and to aid data recovery, the MicroKernel Engine includes one or more committed operations (both transactional and non-transactional) into a bundle of operations called a system transaction. The MicroKernel Engine creates a system transaction bundle for each file. A system transaction can contain operations and user transactions from one or more clients running on the same engine.

Both user transactions and system transactions are atomic. In other words, they happen in such a way that either all or none of the changes occur. If a system failure occurs, the MicroKernel Engine recovers all files involved in the failed system transaction as it reopens the files. All changes made during a failed system transaction (that is, all operations to a file by all clients on that engine since the last completed system transaction) are lost; however, the file is restored to a consistent state, enabling the operations to be attempted again after resolving the cause of the system failure.

Zen guarantees transaction durability for all loggable files except those opened in Accelerated mode. (A file can be logged if it contains at least one unique, or non-duplicatable, key. The key can be system-defined.) Transaction durability is the assurance that before the MicroKernel Engine returns a successful status code to the client for an End Transaction it finishes writing to the transaction log. When you open a file in Accelerated mode, the MicroKernel Engine does not log the file; therefore, the MicroKernel Engine does not log entries to the file. Therefore, the MicroKernel Engine cannot guarantee transaction durability for that file.

After the MicroKernel Engine rolls back a file's system transaction, it replays the log when it next opens the file. Doing so restores those committed operations that were stored in the log but were not written to the file because of the system transaction rollback.

Each system transaction consists of two phases: preparation and writing.

During the preparation phase, the MicroKernel Engine executes all operations in the current system transaction, but writes no pages to the files. The MicroKernel Engine reads uncached pages from the files as needed and creates new page images only in cache.

Any of the following actions can trigger the end of the preparation phase, which marks the beginning of the writing phase:

During the writing phase, the MicroKernel Engine writes to disk all pages prepared in the preparation phase. It first writes all data, index, and variable pages. These are actually shadow pages. While these are being written, the consistency of the files on disk remains the same.

However, the critical part of the system transaction occurs while the PAT pages are being written, because they point to the shadow page as the current page. To protect this phase, the MicroKernel Engine writes a flag in the FCR. When all PAT pages are written, the final FCR is written and the file is now consistent. If a system failure happens during this phase, the MicroKernel Engine recognizes it the next time the file is opened and rolls the file back to the previous state. Then, all durable user transactions in the transaction log file will be implemented in the file.

Doing system transactions less often provides a performance benefit to most configurations. These include client/server, single engine workstation and multi-engine workstation environments where files are opened exclusively.

When the MicroKernel Engine initiates system transactions less often, dirty pages, or pages that need to be written, stay in memory longer. If the application is doing a lot of change operations, these pages might get updated multiple times before being written to disk. This means fewer disk writes. In fact, the most efficient engine is one that writes only when it must.

There are three limits that, when reached, can cause the system transactions to be initiated: Operation Bundle Limit, Cache Size, and Initiation Time Limit. When any of these limits are reached, the MicroKernel Engine initiates a system transaction. See Advanced Operations Guide for more information about these settings.

The best way to do system transactions less often is to set the Operation Bundle Limit and Initiation Time Limit to higher values. You can also increase the size of the cache.

A disadvantage of causing the MicroKernel Engine to perform fewer system transactions is more data in machine memory at any point in time that needs to be written to disk. If a system failure occurs, such as a power outage, more data is lost. Although the MicroKernel Engine is designed to keep files in a consistent usable state, that state may not include the most recent changes. Of course, the use of user transactions with transaction durability will minimize this risk. You should carefully consider the risks of decreasing the frequency of system transactions versus the performance gains.

For example, if your application is using a workstation engine to update a remote file over a slow or non-reliable network connection, you should perform system transactions often so that changes are put onto disk at soon as possible.

The cache is an area of memory the MicroKernel Engine reserves for buffering the pages that it reads. When an application requests a record, the MicroKernel Engine first checks the cache to see if the page containing that record is already in memory. If so, the MicroKernel Engine transfers the record from the cache to the application's data buffer. If the page is not in the cache, the MicroKernel Engine reads the page from the disk into a cache buffer before transferring the requested record to the application. The MicroKernel Engine cache is shared by local clients and used across multiple operations.

If every cache buffer is full when the MicroKernel Engine attempts to transfer a new page into memory, a least-recently-used (LRU) algorithm determines which page in the cache the MicroKernel Engine overwrites. The LRU algorithm reduces processing time by keeping the most recently referenced pages in memory.

When the application inserts or updates a record, the MicroKernel Engine first makes a shadow image of the corresponding page, modifies the page in the cache, and then writes that page to disk. The modified page remains in the cache until the LRU algorithm determines that the MicroKernel Engine can overwrite the image of that page in cache with a new page.

Generally, a larger cache improves performance because it enables more pages to be in memory at a given time. The MicroKernel Engine enables you to specify the amount of memory to reserve for the I/O cache buffers. To determine this amount, consider the application's memory requirements, the total memory installed on your computer, and the combined size of all files that all concurrent Zen applications will access. The configuration setting for this cache is "Cache Allocation Size."

In Zen V8 and later releases, a secondary dynamic L2 cache is also available. The configuration setting for this dynamic cache is "Max MicroKernel Engine Memory Usage." See Advanced Operations Guide for more information on configuring these settings.

Page preallocation guarantees that disk space is available when the MicroKernel Engine needs it. You can enhance the speed of file operations if a data file occupies a contiguous area on the disk. The increase in speed is most noticeable on very large files. For more information about this feature, see Page Preallocation.

Using the extended operations – Get Next Extended (36), Get Previous Extended (37), Step Next Extended (38), Step Previous Extended (39), and Insert Extended (40) – can greatly improve performance. Extended operations can reduce the number of MicroKernel Engine requests by 100 to 1 or more, depending on the application. These operations have the ability to filter the records returned so that records not needed by the application are not sent to it. This optimization technique has the best results in client-server environments that have to send data back and forth over a network.

See Multirecord Operations for detailed information on these operations.

The MicroKernel Engine provides the following features for minimizing disk usage requirements:

When you delete a record, the disk space it formerly occupied is put on a Free Space List. When you insert new records, the MicroKernel Engine uses pages on the Free Space List before creating new variable pages. The Free Space Threshold tells the MicroKernel Engine how much free space must remain on a variable page in order for that page to appear on the Free Space List.

This method of reusing free space eliminates the need to reorganize files to reclaim disk space. Also, the Free Space List provides a means of reducing the fragmentation of variable-length records across several pages. A higher Free Space Threshold reduces fragmentation at the cost of requiring more disk space for the file.

When an index page becomes full, the MicroKernel Engine (by default) automatically creates a new index page and moves some of the values from the full index page to the new index page. Turning on the Index Balancing option lets you avoid creating a new index page every time an existing one becomes full. With index balancing, the MicroKernel Engine looks for available space in sibling index pages each time an index page becomes full. The MicroKernel Engine then rotates values from the full index page onto the pages that have space available.

Index balancing increases index page utilization, results in fewer pages, and produces an even distribution of keys among nodes on the same level, thus enhancing performance during read operations. However, using this feature also means that the MicroKernel Engine requires extra time to examine more index pages and may require more disk I/O during write operations. Although the exact effects of balancing indexes vary in different situations, performance on write operations typically degrades by about 5 to 10 percent if you use index balancing.

Index Balancing impacts the performance of a "steady-state" file by making the average change operation a little slower while making the average get operation faster. It does this by fitting more keys in an average index page. A normal index page may be 50 to 65 percent full where an index balanced page is 65 to 75 percent full. This means there are less index pages to search.

If you create indexes with Create Index (31), the index pages will be nearly 100 percent full, which optimizes these files for reading not writing.

With data compression, the MicroKernel Engine compresses file records before inserting or updating them and uncompresses the records when it retrieves them. Because the final length of a compressed record cannot be determined until the record is written to the file, the MicroKernel Engine always designates a compressed file as a variable-record-length file. However, if you use data compression on a fixed-record-length file, the MicroKernel Engine prevents insert and update operations from producing a record that is longer than the fixed-record length specified for the data file.

Because the MicroKernel Engine stores compressed records as variablelength (even if you created the file as not allowing variable-length records), individual records may become fragmented across several data pages if you perform frequent insertions, updates, and deletions. This fragmentation can result in slower access times, because the mMicroKernel EngineWy need to read multiple file pages to retrieve a single record. However, data compression can also result in a significant reduction of the disk space needed to store records that contain many repeating characters. the cMicroKernel EngineWmpresses five or more of the same contiguous characters into 5 bytes.

For more information on this feature, see Record Compression.

Blank truncation conserves disk space. It is applicable only to files that allow variable-length records and that do not use data compression. For more information on this feature, see Blank Truncation.