A portion of the disclosure of this patent document may contain material that is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent files or records, but otherwise reserves all copyright rights whatsoever. The following notice shall apply to this document: Copyright © 2003, Microsoft Corp.
This invention relates to computing, and more particularly to storage and transmission of data objects.
Serialization can be defined as the process of storing the state of an object instance to a storage medium. During this process, the public and private fields of an object and the name of the class, are converted to a stream of bytes, which is then written to a data stream. When an object is subsequently deserialized, an exact clone of the original object may be created.
Consider an object in active computer memory, for example, an object with data describing a person. The person object has a number of subcomponent members, such as name, address, social security number, phone numbers, spouse, height and weight. While the person's name may be important for a particular application, the height and weight may not be. Thus, the name may remain in active memory where it may be modified, while other fields such as height and weight are evicted from active memory to make room for other data. Ultimately, the person object may no longer be needed by the application, and it may be persisted or transmitted to another computer. To persist or transmit an object, the object must be serialized, which refers to formatting an object in a useful, retrievable way.
In the example above, the members of an object, such as the person object, are generally uniform for all objects of the same class. Each person object, for example, has the name, address, social security number, phone numbers, spouse, height and weight members. The information changes from person to person, and for some people the information may be unavailable (“null”), but the existence of the same member fields is generally present for all person objects of the person class. As such, a person class may be thought of as the generic person object. A person object is one instance of a person class. This concept of a class and an instance of a class exists in many programming languages. Regardless of the programming language involved, serialization is typically performed on instances of a class, generating serialized objects.
Objects may comprise members with various types of data. The members may be primitive or complex. Examples of primitive members are “string” such as the name member from the person object, which is a string of letters; and “integer,” such as the social security number from the person object, which is an integer. Examples of complex members are “collection,” such as the phone numbers member, which comprises more than one primitive—in this case, more than one integer; “nested,” which is a member that has some structure beyond a simple primitive member, e.g., the collection of phone numbers, or the spouse member, which refers to another person object; and “subtype,” such as a hypothetical “United States address” type that would be a subtype of an address type, and therefore presumably declares additional members such as a U.S. region or U.S. Post Office Box. Members may be described in many different ways, and relate to each other in any number of patterns. Therefore serializing objects such as the person object involves effectively dealing with the various members and the relationships of those members that may be included in the object.
Serialization of objects presents a number of challenges in the industry. Serialized objects should consume as little storage space as possible. If the size of an object is greatly increased when it is serialized, then the storage cost of the object may be too high. Therefore, compact representation is an important aspect of a serialization format.
Serialized objects should also be efficiently instantiated into active memory. If the processing cost of finding and assimilating the various members of a serialized object is high, it will drain valuable processor resources. Likewise, serialization should allow for instantiation and updating of members of an object without the need to instantiate the entire object. Instantiating the entire person object, for example, only to read or update the person's social security number is a waste of active memory resources needed to store the name, phone number, address, etc. when those members are not involved in the operation.
Serialization formats should also support all data types that may be contained in an object. A very basic serialization format might only support primitives, but more sophisticated formats should support complex members such as the nested members, collection members, and subtype members described above. While a serialization format should be optimal for objects with few levels of nesting and inheritance, because most objects have this characteristic, it should also support many levels of nesting and inheritance to ensure that the serialization can be flexibly used for a broad range of classes. A serialization format should also be flexible in handling very large members. Some members may be, for example, a music file, a photograph, or a movie, and such large members pose a challenge in serialization that will be explained in greater detail below.
Previous serialization formats have several notable deficiencies. One such format is known as XML Serialization. XML serialization provides a token for each member. The token comprises metadata that identifies a member, usually a member immediately following the token. Therefore, XML serialization may be visualized as follows:
The problems with such a serialization format are, first, verbosity: the storage of metadata tokens with each and every member consumes a large amount of disk space. Second, retrieval is impaired in such a format, because in order to find a desired member, the tokens must be searched. This may involve a high active memory cost, because the most effective way to read or update an object that is serialized in this manner may be to instantiate the entire object.
Another serialization format is in the “Storage Engine record” format, also referred to as the “SE record,” or simply “record” format. This is an a typical database system record format. In this serialization format, members for objects of a given class are stored in uniformly formatted records. Instead of providing metadata that describes each and every member, there is metadata that describes the contents of all the records for objects of a particular class. This can be visualized as provided in
The SE record serialization format does not require metadata with each individual member, so it is a more compact serialization technique. Instead, it requires access to metadata describing the layout of the members on disk, such as the Metadata for Person Objects table of
Further, various storage formats have been designed to allow users of databases to efficiently store objects within a database. These storage formats can be better supported with a more flexible serialization format. For example, should be distinguished from the serialization format provided herein. For example U.S. patent application Ser. No. 10/692,225, Attorney Docket No. MSFT 2852/306819.01, titled “system and method for object persistence in a database store,” is directed to allowing a user to ‘import’ classes and methods written in an object oriented language like C# into a database. It further allows a user to store C# objects in a database and to invoke methods on the objects. It provides multiple flavors of persistence to a user. A user can define his own serialization format, use Common Language Runtime (“CLR”) serialization (provided by C# language itself), or let the SQL server store an object in its own format. These options, particularly the latter, provide a performance advantage, as MICROSOFT SQL SERVER® can retrieve or update some fields of an object without actually instantiating a C# object. Of course, some operations, such as method invocation, still require instantiation of a C# object.
Similar background and related technology descriptions may be found in U.S. patent application Ser. No. 10/692,227, Attorney Docket No. MSFT—2850/306820.1, titled “System and Method for Storing and Retrieving a Field of a User Defined Type Outside of a Database Store.” This application discusses filestreams in UDTs, which may be serialized according to the techniques described herein. Such advanced database technologies can benefit from a more flexible and higher performance serialization format. Likewise, improved techniques for performing operations on serialized objects would better support such advanced database technologies.
The trade-offs involved in serialization formats are thus metadata on-disk memory overhead of the format, versus active memory overhead of locating a member, versus processing cost of locating a member, versus cost of doing an update, versus flexibility in handling large fields. In light of these trade-offs, there is an ongoing and heretofore unaddressed need in the industry to raise the bar with respect to serialization techniques.
A method and system for fragment-based serialization places one or more members in fragments. Fragments may comprise a header and a payload. A header can provide useful information about the fragment, such as an indication of fragment type and an indication of fragment length. A payload may comprise one or more members of an object. Various fragment types are provided for efficiency and flexibility in storing and retrieving object members. Primitive members may be stored in a fragment with a record format payload. This configuration allows for fast location and updating of primitives. Large Object (“LOB”) members may be stored in fragments that have a field for setting forth location types for locations of LOB and FS members. Collections may be stored in a series of fragments, a first fragment to indicate a start of a collection, one or more second fragments to serialize collection elements, and a terminator fragment to indicate the end of a collection. These and other fragment types may be organized according to rules that govern generating fragments, placing members in fragments, and sequencing fragments in a manner that provides additional functionality to the serialization format.
FIGS. 11(A-H) illustrate various fragment types for use in serializing data in accordance with preferred embodiments of the invention.
Certain specific details are set forth in the following description and figures to provide a thorough understanding of various embodiments of the invention. Certain well-known details often associated with computing technology are not set forth in the following disclosure, however, to avoid unnecessarily obscuring the various embodiments of the invention. Further, those of ordinary skill in the relevant art will understand that they can practice other embodiments of the invention without one or more of the details described below. Finally, while various methods are described with reference to steps and sequences in the following disclosure, the description as such is for providing a clear implementation of embodiments of the invention, and the steps and sequences of steps should not be taken as required to practice this invention.
It is an object of the invention to provide a method and system for improved object serialization, as well as to provide techniques for operations on serialized objects. In this regard a serialization that provides compact representation is provided. Objects serialized in the provided format can be efficiently instantiated into active memory, thereby reducing processing costs of finding and assimilating the various members of a serialized object. Likewise members of objects may be instantiated and updated without the need to instantiate an entire object. Moreover, support for a wide range of data types, including user-defined data types (“UDTs”) is provided. The serialization format can be optimized for objects with few levels of nesting and inheritance but also supports many levels of nesting and inheritance. It is flexible in handling very large members. The invention can provide a serialization format that is suitable for storage of various types in a single column—for example, an instance of an “employee” object that is a subtype of a “person” object can be stored in a column that is provided to store only “person” objects. Finally, the serialization format allows for efficient addition of new members to a type, also called efficient type evolution.
Fragment-based serialization in accordance with various embodiments of the invention can be conceptualized as a hybrid format that employs some of the elements of XML-style serialization, as described in the background, and some of the elements of SE record serialization, also described in the background, in addition to numerous aspects and advantages that are unique to fragment-based serialization itself. In this regard, the members of an object can be placed into fragments. Fragments are displayed in
With reference to
The fragment payload can comprise a member or members of a serialized object, as well as any other data. This payload can employ SE record format-style serialization for the members inside, which allows for speedy retrieval of members in the payload. Such a record format payload is a characteristic of Fragment 1 in
The header portion of a fragment may contain various fields, as illustrated in
Various embodiments of the invention utilize multiple types of fragments for additional versatility in serializing objects. A variety of suggested fragment types is set forth below. Before describing the fragment types, however, consider the motivation for using different fragment types. One motivating factor is to allow for better serialization of the various types of members that may make up an object. Recall from the background section that objects frequently comprise multiple members of different types. These members may be, for example:
The above is a non-exhaustive list of potential member types, and all member types are considered candidates for use with the fragment-based serialization techniques described herein.
Fragment Types
In order to accommodate various types of members that may exist in an object that is serialized, the invention can be used with several fragment types. One or more fragment types may be useful for only one type of member, while other fragment types are useful for multiple member types.
The various fragment types can have different formats that are tailored to the contents of the fragment. In the following discussion, a fragment type is first set forth, followed by a visual depiction of a suggested fragment format for that type. Fragment attributes in italics are optional and depend on the value of the Type column. The invention is not limited to the fragment types set forth below. In addition to the fragments provided here, new fragment types may be developed for use in accordance with the general principles of fragment-based serialization provided herein.
Binary Fragment—
Objects represented with in a single binary fragment may be flagged in the Type field as “self-terminating” fragments, eliminating any need to include a Terminator Fragment at the end of a serialized object. This “self-terminator” flag may be in the form of a self-terminator bit in the Type field of a fragment. Such a self-terminator bit could also be located in any other field of a fragment header, or in a fragment payload. Objects that are represented by multiple fragments need not set the self-terminator bit, because a Terminator Fragment can be generated to mark the end of the serialized object.
The Length field is optimally 2 bytes, although the length may vary as described above. The Length field may be used to indicate a length of a payload. The payload in a Binary Fragment may contain any data. In preferred embodiments, the payload contains all primitive members of an object. The payload in such a fragment can be an SE record, to allow for efficient cracking and updating of the primitives or other members stored therein.
LOB Fragment—
Information stored in the Value Type field can describe a type of location in which a LOB is stored. Allowing additional description of the location of LOB members provides flexibility in handling large values. When LOB data (in contrast to a LOB reference) is stored in LOB fragment, an application (or user of a computer) can initiate the generation of a LOB Inlined type fragment, use an 8 byte length, and put the LOB inline. In other words, the LOB can be placed in the payload of the LOB Fragment. If the Value Type field indicates a LOB Inlined type, the length field can be, for example, 8 bytes and the payload can contain a LOB value.
It may not always be desirable to include a LOB inlined with a serialized object. This is because a LOB can take up an large amount of space. Therefore, the Value Type field can indicate a LOB Pointer type, meaning that the payload of the fragment contains a pointer to a LOB location. In this scenario, the Length field can be, for example, 2 bytes, and the payload can contain a LOB reference. The Value Type field can also indicate a LOB Delayed type which can mean that the fragment payload contains a LOB reference to a cell in a database that presumably contains a LOB. In this alternate scenario, the fragment length can be, for example, 2 bytes and the payload can contain a cell reference. A ‘cell reference’ is a combination of a table identifier, a row identifier, and a column identifier. When combined with a ‘path’ (as described below) for a LOB fragment, a cell reference gives enough information to locate actual LOB data. Any other location type information can be contained in the Value Type field of a LOB or FS fragment. Providing such an additional location type field for LOB and FS Fragments gives the serialization format additional flexibility while keeping overhead low.
Note that any of the fragments discussed herein may be null if a particular object does not have a member that is provided for in a serialization for a particular class. If a fragment is null, this information can be set in a bit in the type field for the fragment. In this regard, the Length field and the payload may be removed from the LOB fragment of
FS (File Stream) Fragment—
Terminator Fragment—
Collection Start Fragment—
Collection Element Fragment—
A Locator field may also be included with Collection Element Fragments. The Locator field, like the Value Type field from the LOB and FS fragments, can be used to indicate additional properties of a Collection Element Fragment. For example, Collection Element Fragment may have a payload in SE record format, such as that of the Binary Fragment. The Type field may indicate whether the Collection Element Fragment is a Self-Terminator, by using a bit to indicate whether a fragment terminates itself. If the self-terminator bit is not set, the system can anticipate a Terminator Fragment for the fragment. The Locator field can be used to address a particular element of a collection very much like the GUID of a FS fragment. In the case of a Collection Element Fragment, however the locator field indicates a unique location within a collection and not necessarily a globally unique location.
Regarding the Locator field, it may also be preferable to allow for some anticipation of a locator field in a collection element. A the bit in the Bit field of the Collection Start Fragment may be set to indicate an upcoming Collection Element Fragment with a Locator field. In such a configuration, the system can be configured to deduce that Locator field is present in the Collection Element Fragment.
Null Collection Element Fragment—
Null Fragment—
Annotation and Metadata Fragments—In addition to the other fragments described above, Metadata and Annotation Fragments may be used to describe one or more fragments to a receiver of a serialized object. Such fragments are useful in a variety of situations even though they may not be necessary to deserialize an object. For example, a annotation fragment may allow a client to check information about a particular member or object, or to insert notes or information about a serialized object.
In conclusion, with reference to the descriptions of various fragment types set forth above, one of the advantages attained by an effective serialization format is a reduction in representation overhead. Representation overhead refers to an amount of additional information stored with an object to allow the object to be effectively retrieved. The fragment-based serialization techniques do contain representation overhead, but the overhead is minimized for the corresponding flexibility and functionality of the format.
The first field in a fragment header is the Type field. In preferred embodiments, the Type field consumes 1 byte. This minimizes the associated overhead. In addition, the Locator field associated with the Collection Element Fragment amounts to overhead. Most small unordered collections can be adequately represented in a Locator field that consumes no more than 4 bytes. Larger, ordered collections may consume more than 4 bytes, however, and in this case the Locator field may be replaced with a variable binary (“varbinary”) field that may require more representation overhead. The exact amount of overhead used by the Locator field is considered an implementation detail and is left to the judgment of those skilled in the art, who will appreciate the motivation to reduce representation overhead, but also allow for flexible serialization of collections. Finally, the Length fields associated with several other fragment types (see above) are representation overhead. As described above, the Length field in preferred embodiments can be either 2 bytes or 8 bytes long, depending on the fragment type. The invention is not limited to the exact number of bytes for such fields, and the parameters set forth herein should be considered useful tips from an experienced practitioner and not hard and fast requirements for the invention itself.
Rules for Placing Members in Fragments
As explained above, the various fragment types may serve a variety of fragment members. To serialize an object, a decision must be made as to which fragment type to use for a particular member. An object containing, for example, the primitive, nested, collection, and subtype members can be decomposed into fragments according to a set of rules. While the invention is not limited to a particular correlation of fragment types and member types, a useful set of rules has been developed and is explained in this section. Note that the rules are for clarity of explanation, not to indicate that they must be carried out in a particular order. In practice, the operations corresponding to the rules below may be carried out simultaneously, with the generation, population, and sequencing of fragments conducted as a processor steps through the members of an object. For an exemplary application of these rules to the serialization of an object, refer to
Generate fragments. Embodiments of the invention can be said to engage in type-based container relative fragment generation. In other words, there may be a fragment for each of the following:
Additional fragments may be generated to suit the needs of a particular class. Likewise, the above fragments may not be required for serialization of some classes.
In general, an object may be translated into a fragment-based serialization fragment using a top-to-bottom technique. First, any base type members in an object may be serialized, followed by subtypes. At every nesting level, a scan may be made of the contained members to determine if there are any nested types, subtypes or LOB/FS type members.
Generating Fragments for Primitive Members. Some or all primitive members may be placed into one or more Binary Fragments. A preferred embodiment handles objects with no nested members differently from objects with nested members. These two scenarios are depicted in
With reference to
Generating Fragments for Collections. A flowchart of the process for generating collection fragments is provided in
A collection element can be serialized into of one or more fragments. If a collection element is represented using more than one fragment the representation may have its own Terminator Fragment. The first fragment of a collection element may include a Locator field. One purpose of such a filed can be to keep track of a number of elements processed when serializing a collection. By incrementing a counter in a Locator field to properly indicate a current element for serialization, the serialization process can return to the proper location for serializing the next element of a collection.
Generating LOB and FS Fragments. A flowchart illustrating the generation of fragments for LOB and FS members is provided in
Generating Subtype Fragments. Subtype Fragments may be generated in the same manner as the fragments for any other non-primitive member. In other words, if the Subtype comprises a collection member, a Collection Start Fragment along with any Collection Element Fragments and a Terminator Fragment may be generated to mark the end of the serialization of the subtype member. If the subtype member is a nested LOB member, a LOB Fragment may be generated to contain the subtype member. A binary fragment is generated for the subtype, which includes all small primitive members in the subtype, such as int, float etc, in a similar manner as binary fragment is generated for the base type.
Generating Other Fragments. The techniques described herein may be extrapolated to the generation of any other fragments that may be desired or required for the serialization of a particular object or class of objects.
Populate fragments. While the fragments may actually be generated and populated at the same time, it is useful for purposes of explaining this aspect of the invention to include an overall plan for how fragments are populated. This overall plan is provided in
Sequence of Fragments. Fragments are stored in a sequence that makes up a serialization for any particular class. If a class of objects contains only primitive members, the objects can be serialized into a self-terminating single fragment. Classes with complex or other non-primitive members, can be serialized into more than one fragment. If a serialization for a class comprises more than one fragment, then one or more Terminator Fragments can also be generated. The set of fragments from the starting fragment to the final Terminator Fragment comprises the sequence of fragments as that term is used in this description.
Fragments corresponding to subtypes in an instance may be nested under a fragment for the base type. Using the tPerson, tEmployee and tPartTimeEmployee example from
Example Serialization
Having set forth a number of potential member types, a number of fragment types, and an exemplary set of rules for placing the members in fragments, an example serialization using an embodiment of the serialization format described herein will prove instructive. In this regard,
Referring to
An instance of tEmployee with a non-empty collection of addresses can have the following fragments in the following order:
With reference to the above exemplary fragment sequences, note that for the purpose of supporting reuse of Storage Engine record building/cracking code, as may be desired in those embodiments that make use of a record format in one or more fragments, it is preferable to require that no level of a class for use with the fragment-based serialization techniques described herein may exceed 7 kilobytes (7k) in size. In the example given above, tperson (excluding the associated collection of addresses) should be less than 7 kilobytes in size. Likewise, each taddress should be less than 7k. Note that while this restriction is disclosed for compliance with statutory requirements for indicating the best mode of practicing the invention, allowing a level to exceed 7k is considered a feasible, though less desirable implementation of the invention. This restriction can be enforced at execution time by a serialization library (“SL”). As will be understood, the serialization is responsible for recognizing and parsing the various fragments.
Note, from the example above, that by placing members in one or more fragments, the task of locating a member involves locating the fragment or fragments in which the member has been placed. If there is more than one such fragment, the first fragment may be located first. In this-regard, the position of a fragment relative to the first fragment of a serialization can be determined from metadata. Note that while this technique does not provide for direct addressability of fragments, the task of comparing tokens for each member to identify which aspect of the class an associated member represents is eliminated. Instead, the serialization metadata, when scanned, can quickly direct the processor to an appropriate fragment.
Another advantage of the invention as illustrated in the above exemplary serialization is that serialization of an object can be accomplished in one pass. A serializer process may proceed in a top-down manner, from base type to subtype and containing type to nested type. At each level of nesting, the serializer process may produce one or more fragments. Such a serializer is never required to update a previously generated fragment, though it may be configured to do so if desired.
It will be acknowledged that the serialization format set forth herein can be configured to support fixed, variable length and bit type primitive members; objects nested within other objects; inheritance; inline and out-of-line LOBs (including file streams); ordered collections (providing via the locator function a serialization of collection elements in the appropriate order); unordered collections; and null values. It may also support composable serializations. In this regard, a nested object may be extracted from an existing serialization without a trace of the fragment container in which the object was serialized, i.e., without any state corresponding to its container. This property of embodiments of the invention allows insertion or updating of entire nested objects.
Note also, from the example serialization above, that the proposed serialization format supports adding members to a class without updating any other existing data. The addition of fields to a class is very common in the use of XML schemas, where schemas are frequently updated by adding fields. In this regard, a fragment of any type can be altered to add new members (primitive, collections, or nested objects) as long as they have a default value of NULL and are added to the end of an existing type. And the addition of large members to an object need not be of concern as with record format serialization, because objects stored in the proposed fragment-based format can be stored as a “bag of bytes,” in a single column of a database, and can be arbitrarily large.
Yet another advantage of the serialization format as embodied in the example serialization above is that the identity of a fragment need not be stored with the fragment. Instead, the path to a fragment can be used to reveal fragment identity. The path can be determined from the type metadata of any given object that is operated on. A path identifies a particular fragment in a serialization. Path can be, for example a set of numbers at each fragment identifying a fragment at each nesting level and subtype level. Since fields of an object reside in a pre-determined fragment, they can be located using the path. In a sense, paths are like addresses for fragments. A path can be stored along with a fragment or can be computed by navigating from a first serialized fragment. In order to support embedding and nesting of types a path may account for nesting levels and subtype levels.
The fragment-based serialization format also allows for supporting access to members without instantiating the object. Using the path as a fragment identifier, as explained above, the a member location process, or navigator, can navigate to any desired fragment. Once positioned at a fragment, such a process can allow access to the fragment itself or to an entire sequence of fragments rooted at the located fragment. By providing a directory to a serialization in the form of a map to each fragment, locating a fragment can be accomplished even faster. Such a directory can store fragments in a table organized as a BTree. In such an implementation, one fragment may be stored per row, allowing the path to the fragment to be used as part of a key for each row.
A navigator can also efficiently skip fragments that are not of interest for a particular operation. Navigation of the fragments serialized according to the invention may comprise tracking either a number of nesting levels that are open, or a subtype number. Once the navigator arrives at a desired nesting level or subtype, the navigator may count the number of fragments at that level or subtype. Such a fragment could itself start a new level of nesting.
Further to the advantages of locating members of objects which are serialized in a fragment-based format, access to primitive members can comprise the following simple operation. First, a navigator can locate a Binary Fragment. Primitive members, as described above, are advantageously stored in such a fragment. Next, the required member can be extracted using standard, optimized record cracking code. It can also be updated using standard, optimized record building code. This simple operation provides for high-performance location of primitive members in a serialized object that can conveniently be stored in a single database column. An advantage of allowing for member location in this fashion is that, as mentioned above, members may be updated without the need to instantiate an entire host object. This allows for the replacement of a fragment or sequence of fragments without instantiating an entire object. Each fragment can be self-contained, the invention can be configured so the identifier of a fragment sequence is the presence of a begin-nesting fragment and Terminator Fragment. This allows for performance of updates without fixing up lengths anywhere else, and avoids the offset tables of standard record-format serialization.
Storage of a stream of fragments as generated according to the techniques set forth herein may be conducted by storing a stream of fragments as a LOB. Such a LOB may have a tree-structured storage format whose key is an offset location. This technique for storing the stream of fragments provides for a predictable insert and delete time depending on the size of the LOB. It also allow for updating only portions of the LOB. The on-wire format for objects serialized in fragments is the same as the on-disk format with respect to the shape of the fragment headers. In the case of fragments other than the LOB and FS fragments, there need be no change at all for fragments that are on-wire versus on-disk. This aspect of the invention allows objects to be fast-copied onto a wire, providing significant gains in the speed of transporting objects from location to location. Note, however, that fragment contents may vary for LOB and FS fragments to provide for additional flexibility. This aspect of the invention is explained above with the summary of potential fragment types.
Operations on Fragment-Serialized Objects
The simplicity, efficiency, and flexibility of performing operations on an object that is serialized in a given format are effective criteria in evaluating the performance of the serialization format. Embodiments of the invention as described herein are characterized by significant gains in this regard. This may be especially true when operations are performed using the techniques provided in this section. Note, however, that the following list of operations are not considered to be an exhaustive list of possible operations on objects that are serialized according to the techniques of the invention, nor are the descriptions of operation techniques the exclusive manner of performing such operations. The following operations are numbered for ease of reference, and not to indicate any order or sequence in performing the operations. Rather, each of the operations may be performed independently or in conjunction with any other operations.
One of the advantages of the present invention is that it allows for storage of an object in a single column of a table in a database, as in
The following exemplary operations will be explained with reference to a generic User-Defined Type (“UDT”) object, and a specific implementation of the invention designed for serialization of such an object. Basic algorithms for operations that may be performed on a UDT object stored as fragment stream are provided.
Operation 1: Uniquely identifying an object using its path. When a UDT is stored as a sequence of fragments, every fragment in the sequence can be uniquely identified by a ‘path’. A path is a sequence of steps, where each step can be exactly one of:
Operation 2, Implementing a fragment stream over a byte stream. Databases have been supporting character/binary data of bounded and unbounded lengths for many years. Providing byte stream interfaces on top of character/binary data is a known technology. Byte stream interface includes methods such as
Note that above list is intended to be representative but not exhaustive. As mentioned before, a UDT is stored as a fragment stream. The fragment stream can be implemented on top of a byte stream.
Operation 3: Locating a fragment from path information. Consider a fragment stream representing a UDT. This section explains how to locate a fragment in it given its path.
If the path is valid, the fragment corresponding to the path must be found, except when a null UDT is encountered half-way through, in which case we can not traverse further and the fragment can be considered not found. For example, when looking for ‘manager’ field inside a ‘department’ UDT, if the fragment for ‘department’ object itself is a null fragment, then this method returns FALSE to indicate that the fragment for ‘manager’ is not found.
Note that schema evolution introduces another situation where the fragment corresponding to a path can be missing even if the path is valid. Here is the basic algorithm of LocateFragment. Initially, currentPath in the FragmentStream object is an empty path, and the current fragment is the first fragment.
Boolean
FragmentStream::AdvanceTillStep( ). Further to locating a fragment, the “advance till step” operation may also be used. This method compares two steps according to the following rules. First, a ‘nesting step’ is less than another ‘nesting step’ if fragment ID in the former is less than that in the later. Second, an ‘inheritance step’ is less than another ‘inheritance step’ if depth in the former is less than that in the latter. Third, a ‘locator step’ is less than another ‘locator step’ if the locator in the former is less than that in the latter. Fourth, a ‘nesting step’ is always less than ‘inheritance step’ (as we store all nested fields before the fields from subtypes). Finally, a ‘locator step’ is incomparable with ‘nesting step’ or an ‘inheritance step’ and it is an error to attempt such a comparison. For example, refer to the following algorithm:
FragmentStream::GetNextFragment( ). Further, a get next fragment operation may be implemented as follows:
Operation 4: Selecting a primitive field. Selection of a primitive field involves first locating the fragment containing the primitive field using its path. The payload of this fragment is in record format and has the primitive field. Next, one can use standard optimized record manipulation code to extract the primitive field from the payload.
Operation 5: Updating a primitive field. Updating a primitive field can be accomplished in three steps: First, locate the fragment containing the primitive field using its path in the same way as selecting a primitive field. Next, make a copy of the payload of the fragment and use standard optimized record manipulation code to replace the old value of the primitive field that needs to be updated with the new value. This gives a new payload, which could be longer or shorter than the original payload. Third, update the fragment by replacing old payload with the new payload. Note that this may increase or decrease the length of fragment and its length will have to adjusted accordingly.
Operation 6: Copying an entire embedded UDT. Copying out an entire embedded UDT involves first using locateFragment( ), set forth above, to locate the fragment that marks the beginning of the embedded UDT from a path of the embedded UDT. Second, one can use a CopyOutFragmentSequence( ) function explained below to copy out the fragments belonging to the embedded UDT.
Further to copying a UDT, a FragmentStream::CopyOutFragmentSequence( ) function may be employed, using the following basic algorithm:
Still further, a continue navigation function “FragmentStream::CanContinue Navigation( )” can follow this basic algorithm:
Operation 7: Removing all fragments of a embedded UDT. Similar to CopyOutFragmentSequence( ), a method DeleteFragmentSequence( ) is provided to delete all the fragments belonging to an embedded UDT. For each fragment, the underlying ByteStream class can be used to remove the bytes for that fragment. Special processing is needed if the fragment is filestream fragment to actually delete the underlying file.
Operation 8: Updating an embedded UDT with a new UDT. An algorithm referred to by the inventors as “FragmentStream::ReplaceFragmentSequence( )” can first locate an embedded UDT using its path, next use DeleteFragmentSequence( ) to delete all the fragments that belong to the embedded UDT, and finally read the fragments for the new UDT and insert them at the current location. Again, the underlying ByteStream class can be used to put the bytes for the new fragment. Special processing is needed to insert filestream fragment as the file needs to be created with the proper data, and pointer to that file needs to be put in the fragment.
Operation 9: Inserting a collection member. Consider inserting a UDT object A, supplied in the form a FragmentStream object, as a member of a collection in a FragmentStream object representing object B. The following algorithm is referred to by the inventors as FragmentStream::InsertCollectionElement( ):first, locate the collection in the B using path to the collection. Next, find a locator and position of insert. As mentioned in earlier, if a collection member was deleted earlier, it results in a ‘gap’ in the locators. If any such gap exists, the unused locator is assigned to the new member and it is inserted such that all members are laid out in increasing order of locators. Otherwise, the new member is assigned a locator that is 1 more than last member and is inserted after the last member. Note that the first fragment of A will have to be modified to put the locator in it.
Operation 10: Deleting a collection member. An algorithm referred to by the inventors as “FragmentStream::DeleteCollectionElement( )” can specify a member to delete is using its locator. Deletion involves first locating the collection using its path, and next locating the member to delete within the collection. Note that this is a simpler problem that locating a fragment given its path, because we are just dealing with locator step here. So similar logic as in LocateFragment method can be used for this. Next, call DeleteFragmentSequence once we are positioned on the first fragment belonging to the member to delete.
Operation 11: Updating an entire collection or a single collection member. Replacement of an entire collection with a different collection is done in the same way as replacing an embedded UDT with another UDT. Similarly, once a collection member has been located, updating it can be done in the same way as updating an embedded UDT.
Operation 12: Selecting or updating multiple fields of a UDT. When selecting or updating multiple fields, the following optimizations are performed: First, ordering the selects/updates. Comparison and ordering of two steps has been explained above. Path can be considered as a string of steps and a ‘lexicographical’ ordering can be defined on paths. Refer to the following algorithm:
Fields are selected or updated in the increasing order of the paths to the fragments containing them. Note that this ordering is the same order in which the fragments appear in the fragment stream. Multiple primitive fields could be located in the same fragment. In such a case, once the fragment containing them is visited, the standard record manipulation code is used to select or update all desired fields efficiently, without the need to visit the fragment again. By doing the above, even in a complex UDT, at most one pass is required over its fragment stream to select or update all the desired fields of that UDT.
Operation 13: Enhancing LocateFragment to make use of current position. The previous depiction of LocateFragment algorithm always started at the beginning of fragment stream. However, LocateFragment needs to start from the beginning only for locating the fragment containing the first field that needs to be selected or updated. For locating the fragment containing subsequent fields that need to be selected or updated, locateFragment can start from the current position. Enhancements needed to LocateFragment are briefly mentioned below. There are two basic cases here:
Operation 14: Lazy Materialization. When sending a serialization of a UDT from server to client, sending LOB data and file data from filestream fields tend to be the most time consuming factor. So the fragment stream manager can provide a ‘lazy materialization’ option.
When the serialization of a UDT is requested with lazy materialization option, ‘cookies’ are returned in place of LOB/filestream data. Caller can subsequently ask for the full LOB/filestream data by passing the path of the LOB/filestream fragment and the ‘cookie’. The path and the cookie will give enough information for the fragment stream manager to retrieve the LOB/filestream data.
Operation 15: Schema evolution. Schema evolution refers to changing the UDT by adding, removing, or modifying fields (such as changing data type of a field) of a UDT, or changing the inheritance hierarchy by defining new UDTs etc. Such changes may impact instances of UDT that have already been persisted. A simple solution is to ensure that any field IDs and fragment IDs assigned to existing fields continue to remain the same when new fields are added. Then, addition of new fields to a UDT can be supported without modifying any pre-existing persisted instances of the UDT.
Finally, it should be understood that the various techniques described herein may be implemented in connection with hardware or software or, where appropriate, with a combination of both. Thus, the methods and apparatus of the present invention, or certain aspects or portions thereof, may take the form of program code (i.e., instructions) embodied in tangible media, such as floppy diskettes, CD-ROMs, hard drives, or any other machine-readable storage medium wherein, when the program code is loaded into and executed by a machine, such as a computer, the machine becomes an apparatus for practicing the invention. In the case of program code execution on programmable computers, the computing device generally includes a processor, a storage medium readable by the processor (including volatile and non-volatile memory and/or storage elements), at least one input device, and at least one output device. One or more programs that may implement or utilize the user interface techniques of the present invention, e.g., through the use of a data processing API, reusable controls, or the like, are preferably implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language, and combined with hardware implementations.
Although exemplary embodiments refer to utilizing the present invention in the context of one or more stand-alone computer systems, the invention is not so limited, but rather may be implemented in connection with any computing environment, such as a network or distributed computing environment. Still further, the present invention may be implemented in or across a plurality of processing chips or devices, and storage may similarly be effected across a plurality of devices. Such devices might include personal computers, network servers, handheld devices, supercomputers, or computers integrated into other systems such as automobiles and airplanes. Therefore, the present invention should not be limited to any single embodiment, but rather should be construed in breadth and scope in accordance with the appended claims.
An alternate layout for fragment validation code is described below, where the grammar of a serialization is illustrated in Backus Naur Form (“BNF”).
A symbol in quotes “ ” is a terminal. For example “self-terminating bin frag” represents a terminal. The various terminal symbols are defined in the previous section.
A symbol in angle brackets < > represents a non-terminal. For example, <containment frag> is a non-terminal.
A symbol in curly braces { } indicates that symbol may be repeated zero or more times. For example, {<containment frag>} indicates there may be zero or more instances of <containment frag>.
Annotation fragments can exist anywhere in the stream.