HDF5: Disk Format Implementation

  1. Disk Format Level 0 - File Signature and Boot Block
  2. Disk Format Level 1 - File Infrastructure
    1. Disk Format Level 1A - B-link Trees
    2. Disk Format Level 1B - Symbol Table
    3. Disk Format Level 1C - Symbol Table Entry
    4. Disk Format Level 1D - Local Heaps
    5. Disk Format Level 1E - Global Heap
    6. Disk Format Level 1F - Free-Space Index
  3. Disk Format Level 2 - Data Objects
    1. Disk Format Level 2a - Data Object Headers
      1. Name: NIL
      2. Name: Simple Data Space
      3. Name: Data-Space
      4. Name: Data-Type
      5. Name: Data Storage - Fill Value
      6. Name: Reserved - not assigned yet
      7. Name: Data Storage - Compact
      8. Name: Data Storage - External Data Files
      9. Name: Data Storage - Layout
      10. Name: Reserved - not assigned yet
      11. Name: Reserved - not assigned yet
      12. Name: Data Storage - Filter Pipeline
      13. Name: Attribute
      14. Name: Object Name
      15. Name: Object Modification Date & Time
      16. Name: Shared Object Message
      17. Name: Object Header Continuation
      18. Name: Symbol Table Message
    2. Disk Format: Level 2b - Shared Data Object Headers
    3. Disk Format: Level 2c - Data Object Data Storage

Disk Format Implementation

The format of a HDF5 file on disk encompasses several key ideas of the current HDF4 & AIO file formats as well as addressing some short-comings therein. The new format will be more self-describing than the HDF4 format and will be more uniformly applied to data objects in the file.

Three levels of information compose the file format. The level 0 contains basic information for identifying and "boot-strapping" the file. Level 1 information is composed of the object directory (stored as a B-tree) and is used as the index for all the objects in the file. The rest of the file is composed of data-objects at level 2, with each object partitioned into header (or "meta") information and data information.

The sizes of various fields in the following layout tables are determined by looking at the number of columns the field spans in the table. There are three exceptions: (1) The size may be overridden by specifying a size in parentheses, (2) the size of addresses is determined by the Size of Addresses field in the boot block, and (3) the size of size fields is determined by the Size of Sizes field in the boot block.

Disk Format: Level 0 - File Signature and Boot Block

The boot block may begin at certain predefined offsets within the HDF5 file, allowing a block of unspecified content for users to place additional information at the beginning (and end) of the HDF5 file without limiting the HDF5 library's ability to manage the objects within the file itself. This feature was designed to accommodate wrapping an HDF5 file in another file format or adding descriptive information to the file without requiring the modification of the actual file's information. The boot-block is located by searching for the HDF5 file signature at byte offset 0, byte offset 512 and at successive locations in the file, each a multiple of two of the previous location, i.e. 0, 512, 1024, 2048, etc.

The boot-block is composed of a file signature, followed by boot block and object directory version numbers, information about the sizes of offset and length values used to describe items within the file, the size of each object directory page, and a symbol table entry for the root object in the file.

HDF5 Boot Block Layout
byte byte byte byte

HDF5 File Signature (8 bytes)

Version # of Boot Block Version # of Global Free-Space Storage Version # of Object Directory Reserved
Version # of Shared Header Message Format Size of Addresses Size of Sizes Reserved (zero)
Symbol Table Leaf Node K Symbol Table Internal Node K
File Consistency Flags
Base Address
Address of Global Free-Space Heap
End of File Address
Reserved Address

Root Group Symbol Table Entry

Field Name Description
File Signature This field contains a constant value and can be used to quickly identify a file as being an HDF5 file. The constant value is designed to allow easy identification of an HDF5 file and to allow certain types of data corruption to be detected. The file signature of a HDF5 file always contain the following values:

decimal 137 72 68 70 13 10 26 10
hexadecimal 89 48 44 46 0d 0a 1a 0a
ASCII C Notation \211 H D F \r \n \032 \n

This signature both identifies the file as a HDF5 file and provides for immediate detection of common file-transfer problems. The first two bytes distinguish HDF5 files on systems that expect the first two bytes to identify the file type uniquely. The first byte is chosen as a non-ASCII value to reduce the probability that a text file may be misrecognized as a HDF5 file; also, it catches bad file transfers that clear bit 7. Bytes two through four name the format. The CR-LF sequence catches bad file transfers that alter newline sequences. The control-Z character stops file display under MS-DOS. The final line feed checks for the inverse of the CR-LF translation problem. (This is a direct descendent of the PNG file signature.)
Version # of the Boot Block This value is used to determine the format of the information in the boot block. When the format of the information in the boot block is changed, the version # is incremented to the next integer and can be used to determine how the information in the boot block is formatted.
Version # of the Global Free-Space Storage This value is used to determine the format of the information in the Global Free-Space Heap. Currently, this is implemented as a B-tree of length/offset pairs to locate free space in the file, but future advances in the file-format could change the method of finding global free-space. When the format of the information is changed, the version # is incremented to the next integer and can be used to determine how the information is formatted.
Version # of the Object Directory This value is used to determine the format of the information in the Object Directory. When the format of the information in the Object Directory is changed, the version # is incremented to the next integer and can be used to determine how the information in the Object Directory is formatted.
Version # of the Shared Header Message Format This value is used to determine the format of the information in a shared object header message, which is stored in the global small-data heap. Since the format of the shared header messages differ from the private header messages, a version # is used to identify changes in the format.
Size of Addresses This value contains the number of bytes used for addresses in the file. The values for the addresses of objects in the file are relative to a base address, usually the address of the boot block signature. This allows a wrapper to be added after the file is created without invalidating the internal offset locations.
Size of Sizes This value contains the number of bytes used to store the size of an object.
Symbol Table Leaf Node K Each leaf node of a symbol table B-tree will have at least this many entries but not more than twice this many. If a symbol table has a single leaf node then it may have fewer entries.
Symbol Table Internal Node K Each internal node of a symbol table B-tree will have at least K pointers to other nodes but not more than 2K pointers. If the symbol table has only one internal node then it might have fewer than K pointers.
Bytes per B-Tree Page This value contains the # of bytes used for symbol pairs per page of the B-Trees used in the file. All B-Tree pages will have the same size per page.
(For 32-bit file offsets, 340 objects is the maximum per 4KB page, and for 64-bit file offset, 254 objects will fit per 4KB page. In general, the equation is:
<# of objects> = FLOOR((<page size>-<offset size>)/(<Symbol size>+<offset size>))-1 )
File Consistency Flags This value contains flags to indicate information about the consistency of the information contained within the file. Currently, the following bit flags are defined: bit 0 set indicates that the file is opened for write-access and bit 1 set indicates that the file has been verified for consistency and is guaranteed to be consistent with the format defined in this document. Bits 2-31 are reserved for future use. Bit 0 should be set as the first action when a file is opened for write access and should be cleared only as the final action when closing a file. Bit 1 should be cleared during normal access to a file and only set after the file's consistency is guaranteed by the library or a consistency utility.
Base Address This is the absolute file address of the first byte of the hdf5 data within the file. Unless otherwise noted, all other file addresses are relative to this base address.
Address of Global Free-Space Heap This value contains the relative address of the B-Tree used to manage the blocks of data which are unused in the file currently. The free-space heap is used to manage the blocks of bytes at the file-level which become unused with objects are moved within the file.
End of File Address This is the relative file address of the first byte past the end of all HDF5 data. It is used to determine if a file has been accidently truncated and as an address where file memory allocation can occur if the free list is not used.
Reserved Address This address field is present for alignment purposes and is always set to the undefined address value (all bits set).
Root Group Symbol Table Entry This symbol-table entry (described later in this document) refers to the entry point into the group graph. If the file contains a single object, then that object can be the root object and no groups are used.

Disk Format: Level 1A - B-link Trees

B-link trees allow flexible storage for objects which tend to grow in ways that cause the object to be stored discontiguously. B-trees are described in various algorithms books including "Introduction to Algorithms" by Thomas H. Cormen, Charles E. Leiserson, and Ronald L. Rivest. The B-link tree, in which the sibling nodes at a particular level in the tree are stored in a doubly-linked list, is described in the "Efficient Locking for Concurrent Operations on B-trees" paper by Phillip Lehman and S. Bing Yao as published in the ACM Transactions on Database Systems, Vol. 6, No. 4, December 1981.

The B-link trees implemented by the file format contain one more key than the number of children. In other words, each child pointer out of a B-tree node has a left key and a right key. The pointers out of internal nodes point to sub-trees while the pointers out of leaf nodes point to other file data types. Notwithstanding that difference, internal nodes and leaf nodes are identical.

B-tree Nodes
byte byte byte byte
Node Signature
Node Type Node Level Entries Used
Address of Left Sibling
Address of Right Sibling
Key 0 (variable size)
Address of Child 0
Key 1 (variable size)
Address of Child 1
...
Key 2K (variable size)
Address of Child 2K
Key 2K+1 (variable size)

Field Name Description
Node Signature The value ASCII 'TREE' is used to indicate the beginning of a B-link tree node. This gives file consistency checking utilities a better chance of reconstructing a damaged file.
Node Type Each B-link tree points to a particular type of data. This field indicates the type of data as well as implying the maximum degree K of the tree and the size of each Key field.
0
This tree points to symbol table nodes.
1
This tree points to a (partial) linear address space.
Node Level The node level indicates the level at which this node appears in the tree (leaf nodes are at level zero). Not only does the level indicate whether child pointers point to sub-trees or to data, but it can also be used to help file consistency checking utilities reconstruct damanged trees.
Entries Used This determines the number of children to which this node points. All nodes of a particular type of tree have the same maximum degree, but most nodes will point to less than that number of children. The valid child pointers and keys appear at the beginning of the node and the unused pointers and keys appear at the end of the node. The unused pointers and keys have undefined values.
Address of Left Sibling This is the file address of the left sibling of the current node relative to the boot block. If the current node is the left-most node at this level then this field is the undefined address (all bits set).
Address of Right Sibling This is the file address of the right sibling of the current node relative to the boot block. If the current node is the right-most node at this level then this field is the undefined address (all bits set).
Keys and Child Pointers Each tree has 2K+1 keys with 2K child pointers interleaved between the keys. The number of keys and child pointers actually containing valid values is determined by the `Entries Used' field. If that field is N then the B-link tree contains N child pointers and N+1 keys.
Key The format and size of the key values is determined by the type of data to which this tree points. The keys are ordered and are boundaries for the contents of the child pointer. That is, the key values represented by child N fall between Key N and Key N+1. Whether the interval is open or closed on each end is determined by the type of data to which the tree points.
Address of Children The tree node contains file addresses of subtrees or data depending on the node level (0 implies data addresses).

Disk Format: Level 1B - Symbol Table

A symbol table is a group internal to the file that allows arbitrary nesting of objects (including other symbol tables). A symbol table maps a set of names to a set of file address relative to the file boot block. Certain meta data for an object to which the symbol table points can be cached in the symbol table in addition to (or in place of?) the object header.

An HDF5 object name space can be stored hierarchically by partitioning the name into components and storing each component in a symbol table. The symbol table entry for a non-ultimate component points to the symbol table containing the next component. The symbol table entry for the last component points to the object being named.

A symbol table is a collection of symbol table nodes pointed to by a B-link tree. Each symbol table node contains entries for one or more symbols. If an attempt is made to add a symbol to an already full symbol table node containing 2K entries, then the node is split and one node contains K symbols and the other contains K+1 symbols.

Symbol Table Node
byte byte byte byte
Node Signature
Version Number Reserved for Future Use Number of Symbols


Symbol Table Entries


Field Name Description
Node Signature The value ASCII 'SNOD' is used to indicate the beginning of a symbol table node. This gives file consistency checking utilities a better chance of reconstructing a damaged file.
Version Number The version number for the symbol table node. This document describes version 1.
Number of Symbols Although all symbol table nodes have the same length, most contain fewer than the maximum possible number of symbol entries. This field indicates how many entries contain valid data. The valid entries are packed at the beginning of the symbol table node while the remaining entries contain undefined values.
Symbol Table Entries Each symbol has an entry in the symbol table node. The format of the entry is described below.

Disk Format: Level 1C - Symbol-Table Entry

Each symbol table entry in a symbol table node is designed to allow for very fast browsing of commonly stored scientific objects. Toward that design goal, the format of the symbol-table entries includes space for caching certain constant meta data from the object header.

Symbol Table Entry
byte byte byte byte
Name Offset (<size> bytes)
Object Header Address
Symbol-Type
Reserved


Scratch-pad Space (16 bytes)


Field Name Description
Name Offset This is the byte offset into the symbol table local heap for the name of the symbol. The name is null terminated.
Object Header Address Every object has an object header which serves as a permanent home for the object's meta data. In addition to appearing in the object header, the meta data can be cached in the scratch-pad space.
Symbol-Type The symbol type is determined from the object header. It also determines the format for the scratch-pad space. The value zero indicates that no object header meta data is cached in the symbol table entry.
0
No data is cached by the symbol table entry. This is guaranteed to be the case when an object header has a link count greater than one.
1
Symbol table meta data is cached in the symbol table entry. This implies that the symbol table entry refers to another symbol table.
2
The entry is a symbolic link. The first four bytes of the scratch pad space are the offset into the local heap for the link value. The object header address will be undefined.
N
Other cache values can be defined later and libraries that don't understand the new values will still work properly.
Reserved These for bytes are present so that the scratch pad space is aligned on an eight-byte boundary. They are always set to zero.
Scratch-Pad Space This space is used for different purposes, depending on the value of the Symbol Type field. Any meta-data about a dataset object represented in the scratch-pad space is duplicated in the object header for that dataset. Furthermore, no data is cached in the symbol table entry scratch-pad space if the object header for the symbol table entry has a link count greater than one.

The symbol table entry scratch-pad space is formatted according to the value of the Symbol Type field. If the Symbol Type field has the value zero then no information is stored in the scratch pad space.

If the Symbol Type field is one, then the scratch pad space contains cached meta data for another symbol table with the format:

Symbol Table Scratch-Pad Format
byte byte byte byte
Address of B-tree
Address of Name Heap

Field Name Description
Address of B-tree This is the file address for the symbol table's B-tree.
Address of Name Heap This is the file address for the symbol table's local heap that stores the symbol names.

Symbolic Link Scratch-Pad Format
byte byte byte byte
Offset to Link Value

Field Name Description
Offset to Link Value The value of a symbolic link (that is, the name of the thing to which it points) is stored in the local heap. This field is the 4-byte offset into the local heap for the start of the link value, which is null terminated.

Disk Format: Level 1D - Local Heaps

A heap is a collection of small heap objects. Objects can be inserted and removed from the heap at any time and the address of a heap doesn't change once the heap is created. Note: this is the "local" version of the heap mostly intended for the storage of names in a symbol table. The storage of small objects in a global heap is described below.

Local Heaps
byte byte byte byte
Heap Signature
Reserved (zero)
Data Segment Size
Offset to Head of Free-list (<size> bytes)
Address of Data Segment

Field Name Description
Heap Signature The valid ASCII 'HEAP' is used to indicate the beginning of a heap. This gives file consistency checking utilities a better chance of reconstructing a damaged file.
Data Segment Size The total amount of disk memory allocated for the heap data. This may be larger than the amount of space required by the object stored in the heap. The extra unused space holds a linked list of free blocks.
Offset to Head of Free-list This is the offset within the heap data segment of the first free block (or all 0xff bytes if there is no free block). The free block contains <size> bytes that are the offset of the next free chunk (or all 0xff bytes if this is the last free chunk) followed by <size> bytes that store the size of this free chunk.
Address of Data Segment The data segment originally starts immediately after the heap header, but if the data segment must grow as a result of adding more objects, then the data segment may be relocated to another part of the file.

Objects within the heap should be aligned on an 8-byte boundary.

Disk Format: Level 1E - Global Heap

Each HDF5 file has a global heap which stores various types of information which is typically shared between datasets. The global heap was designed to satisfy these goals:

  1. Repeated access to a heap object must be efficient without resulting in repeated file I/O requests. Since global heap objects will typically be shared among several datasets it's probable that the object will be accessed repeatedly.

  2. Collections of related global heap objects should result in fewer and larger I/O requests. For instance, a dataset of void pointers will have a global heap object for each pointer. Reading the entire set of void pointer objects should result in a few large I/O requests instead of one small I/O request for each object.

  3. It should be possible to remove objects from the global heap and the resulting file hole should be eligible to be reclaimed for other uses.

The implementation of the heap makes use of the memory management already available at the file level and combines that with a new top-level object called a collection to achieve Goal B. The global heap is the set of all collections. Each global heap object belongs to exactly one collection and each collection contains one or more global heap objects. For the purposes of disk I/O and caching, a collection is treated as an atomic object.

Global Heap Collection
byte byte byte byte
Magic Number
Version Reserved
Collection Size

Object 1


Object 2


...


Object N


Object 0 (free space)

Field Name Description
Magic Number The magic number for global heap collections are the four bytes `G', `C', `O', `L'.
Version Each collection has its own version number so that new collections can be added to old files. This document describes version zero of the collections.
Collection Data Size This is the size in bytes of the entire collection including this field. The default (and minimum) collection size is 4096 bytes which is a typical file system block size and which allows for 170 16-byte heap objects plus their overhead.
Object i for positive i The objects are stored in any order with no intervening unused space.
Object 0 Object zero, when present, represents the free space in the collection. Free space always appears at the end of the collection. If the free space is too small to store the header for object zero (described below) then the header is implied.

Global Heap Object
byte byte byte byte
Object ID Reference Count
Reserved
Object Data Size

Object Data

Field Name Description
Object ID Each object has a unique identification number within a collection. The identification numbers are chosen so that new objects have the smallest value possible with the exception that the identifier `0' always refers to the object which represents all free space within the collection.
Reference Count All heap objects have a reference count field. An object which is referenced from some other part of the file will have a positive reference count. The reference count for Object zero is always zero.
Reserved Zero padding to align next field on an 8-byte boundary.
Object Size This is the size of the the fields above plus the object data stored for the object. The actual storage size is rounded up to a multiple of eight.
Object Data The object data is treated as a one-dimensional array of bytes to be interpreted by the caller.

Disk Format: Level 1F - Free-Space Index (NOT FULLY DEFINED)

The Free-Space Index is a collection of blocks of data, dispersed throughout the file, which are currently not used by any file objects. The blocks of data are indexed by a B-tree of their length within the file.

Each B-Tree page is composed of the following entries and B-tree management information, organized as follows:

HDF5 Free-Space Heap Page
byte byte byte byte
Free-Space Heap Signature
B-Tree Left-Link Offset

Length of Free-Block #1


Offset of Free-Block #1

.
.
.

Length of Free-Block #n


Offset of Free-Block #n

"High" Offset
Right-Link Offset

The elements of the free-space heap page are described below:
Free-Space Heap Signature: (4 bytes)
The value ASCII: 'FREE' is used to indicate the beginning of a free-space heap B-Tree page. This gives file consistency checking utilities a better chance of reconstructing a damaged file.
B-Tree Left-Link Offset: (<offset> bytes)
This value is used to indicate the offset of all offsets in the B-link-tree which are smaller than the value of the offset in entry #1. This value is also used to indicate a leaf node in the B-link-tree by being set to all ones.
Length of Free-Block #n: (<length> bytes)
This value indicates the length of an un-used block in the file.
Offset of Free-Block #n: (<offset> bytes)
This value indicates the offset in the file of an un-used block in the file.
"High" Offset: (4-bytes)
This offset is used as the upper bound on offsets contained within a page when the page has been split.
Right-link Offset: (<offset> bytes)
This value is used to indicate the offset of the next child to the right of the parent of this object directory page. When there is no node to the right, this value is all zeros.

The algorithms for searching and inserting objects in the B-tree pages are described fully in the Lehman & Yao paper, which should be read to provide a full description of the B-Tree's usage.

Disk Format: Level 2 - Data Objects

Data objects contain the real information in the file. These objects compose the scientific data and other information which are generally thought of as "data" by the end-user. All the other information in the file is provided as a framework for these data objects.

A data object is composed of header information and data information. The header information contains the information needed to interpret the data information for the data object as well as additional "meta-data" or pointers to additional "meta-data" used to describe or annotate each data object.

Disk Format: Level 2a - Data Object Headers

The header information of an object is designed to encompass all the information about an object which would be desired to be known, except for the data itself. This information includes the dimensionality, number-type, information about how the data is stored on disk (in external files, compressed, broken up in blocks, etc.), as well as other information used by the library to speed up access to the data objects or maintain a file's integrity. The header of each object is not necessarily located immediately prior to the object's data in the file and in fact may be located in any position in the file.

Object Headers
byte byte byte byte
Version # of Object Header Reserved Number of Header Messages
Object Reference Count

Total Object-Header Size

Header Message Type #1 Size of Header Message Data #1
Flags Reserved

Header Message Data #1

.
.
.
Header Message Type #n Size of Header Message Data #n
Flags Reserved

Header Message Data #n

Field Name Description
Version # of the object header This value is used to determine the format of the information in the object header. When the format of the information in the object header is changed, the version # is incremented and can be used to determine how the information in the object header is formatted.
Reserved Always set to zero.
Number of header messages This value determines the number of messages listed in this object header. This provides a fast way for software to prepare storage for the messages in the header.
Object Reference Count This value specifies the number of references to this object within the current file. References to the data-object from external files are not tracked.
Total Object-Header Size This value specifies the total number of bytes of header message data following this length field for the current message as well as any continuation data located elsewhere in the file.
Header Message Type The header message type specifies the type of information included in the header message data following the type along with a small amount of other information. Bit 15 of the message type is set if the message is constant (constant messages cannot be changed since they may be cached in symbol table entries throughout the file). The header message types for the pre-defined header messages will be included in further discussion below.
Size of Header Message Data This value specifies the number of bytes of header message data following the header message type and length information for the current message. The size includes padding bytes to make the message a multiple of eight bytes.
Flags This is a bit field with the following definition:
0
If set, the message data is constant. This is used for messages like the data type message of a dataset.
1
If set, the message is stored in the global heap and the Header Message Data field contains a Shared Object message. and the Size of Header Message Data field contains the size of that Shared Object message.
2-7
Reserved
Header Message Data The format and length of this field is determined by the header message type and size respectively. Some header message types do not require any data and this information can be eliminated by setting the length of the message to zero. The data is padded with enough zeros to make the size a multiple of eight.

The header message types and the message data associated with them compose the critical "meta-data" about each object. Some header messages are required for each object while others are optional. Some optional header messages may also be repeated several times in the header itself, the requirements and number of times allowed in the header will be noted in each header message description below.

The following is a list of currently defined header messages:


Name: NIL

Type: 0x0000
Length: varies
Status: Optional, may be repeated.
Purpose and Description: The NIL message is used to indicate a message which is to be ignored when reading the header messages for a data object. [Probably one which has been deleted for some reason.]
Format of Data: Unspecified.
Examples: None.

Name: Simple Data Space

Type: 0x0001
Length: varies
Status: One of the Simple Data Space or Data-Space messages is required (but not both) and may not be repeated.

The Simple Dimensionality message describes the number of dimensions and size of each dimension that the data object has. This message is only used for datasets which have a simple, rectilinear grid layout, datasets requiring a more complex layout (irregularly or unstructured grids, etc) must use the Data-Space message for expressing the space the dataset inhabits.

Simple Data Space Message
byte byte byte byte
Version Dimensionality Flags Reserved
Reserved
Dimension Size #1 (<size> bytes)
.
.
.
Dimension Size #n (<size> bytes)
Dimension Maximum #1 (<size> bytes)
.
.
.
Dimension Maximum #n (<size> bytes)
Permutation Index #1
.
.
.
Permutation Index #n

Field Name Description
Dimensionality This value is the number of dimensions that the data object has.
Flags This field is used to store flags to indicate the presence of parts of this message. Bit 0 (the least significant bit) is used to indicate that maximum dimensions are present. Bit 1 is used to indicate that permutation indices are present for each dimension.
Dimension Size #n (<size> bytes) This value is the current size of the dimension of the data as stored in the file. The first dimension stored in the list of dimensions is the slowest changing dimension and the last dimension stored is the fastest changing dimension.
Dimension Maximum #n (<size> bytes) This value is the maximum size of the dimension of the data as stored in the file. This value may be the special value <UNLIMITED> (all bits set) which indicates that the data may expand along this dimension indefinitely. If these values are not stored, the maximum value of each dimension is assumed to be the same as the current size value.
Permutation Index #n (4 bytes) This value is the index permutation used to map each dimension from the canonical representation to an alternate axis for each dimension. If these values are not stored, the first dimension stored in the list of dimensions is the slowest changing dimension and the last dimension stored is the fastest changing dimension.

Examples

Example #1
A sample 640 horizontally by 480 vertically raster image dimension header. The number of dimensions would be set to 2 and the first dimension's size and maximum would both be set to 480. The second dimension's size and maximum would both be set to 640 .
Example #2
A sample 4 dimensional scientific dataset which is composed of 30x24x3 slabs of data being written out in an unlimited series every several minutes as timestep data (currently there are five slabs). The number of dimensions is 4. The first dimension size is 5 and it's maximum is <UNLIMITED>. The second through fourth dimensions' size and maximum value are set to 3, 24, and 30 respectively.
Example #3
A sample unlimited length text string, currently of length 83. The number of dimensions is 1, the size of the first dimension is 83 and the maximum of the first dimension is set to <UNLIMITED>, allowing further text data to be appended to the string or possibly the string to be replaced with another string of a different size. (This could also be stored as a scalar dataset with number-type set to "string")

Name: Data-Space (Fiber Bundle?)

Type: 0x0002
Length: varies
Status: One of the Simple Dimensionality or Data-Space messages is required (but not both) and may not be repeated.
Purpose and Description: The Data-Space message describes space that the dataset is mapped onto in a more comprehensive way than the Simple Dimensionality message is capable of handling. The data-space of a dataset encompasses the type of coordinate system used to locate the dataset's elements as well as the structure and regularity of the coordinate system. The data-space also describes the number of dimensions which the dataset inhabits as well as a possible higher dimensional space in which the dataset is located within.
Format of Data:
HDF5 Data-Space Message Layout
byte byte byte byte
Mesh Type
Logical Dimensionality

The elements of the dimensionality message are described below:
Mesh Type: (unsigned 32-bit integer)
This value indicates whether the grid is polar/spherical/cartesion, structured/unstructured and regular/irregular.
The mesh type value is broken up as follows:

HDF5 Mesh-Type Layout
byte byte byte byte
Mesh Embedding Coordinate System Structure Regularity
The following are the definitions of mesh-type bytes:
Mesh Embedding
This value indicates whether the dataset data-space is located within another dataspace or not:
<STANDALONE>
The dataset mesh is self-contained and is not embedded in another mesh.
<EMBEDDED>
The dataset's data-space is located within another data-space, as described in information below.
Coordinate System
This value defines the type of coordinate system used for the mesh:
<POLAR>
The last two dimensions are in polar coordinates, higher dimensions are cartesian.
<SPHERICAL>
The last three dimensions are in spherical coordinates, higher dimensions are cartesian.
<CARTESIAN>
All dimensions are in cartesian coordinates.
Structure
This value defines the locations of the grid-points on the axes:
<STRUCTURED>
All grid-points are on integral, sequential locations, starting from 0.
<UNSTRUCTURED>
Grid-points locations in each dimension are explicitly defined and may be of any numeric data-type.
Regularity
This value defines the locations of the dataset points on the grid:
<REGULAR>
All dataset elements are located at the grid-points defined.
<IRREGULAR>
Each dataset element has a particular grid-location defined.

The following grid combinations are currently allowed:

<POLAR-STRUCTURED-REGULAR>
<SPHERICAL-STRUCTURED-REGULAR>
<CARTESIAN-STRUCTURED-REGULAR>
<POLAR-UNSTRUCTURED-REGULAR>
<SPHERICAL-UNSTRUCTURED-REGULAR>
<CARTESIAN-UNSTRUCTURED-REGULAR>
<CARTESIAN-UNSTRUCTURED-IRREGULAR>
All of the above grid types can be embedded within another data-space.

Logical Dimensionality: (unsigned 32-bit integer)
This value is the number of dimensions that the dataset occupies.

HDF5 Data-Space Embedded Dimensionality Information
byte byte byte byte
Embedded Dimensionality
Embedded Dimension Size #1
.
.
.
Embedded Dimension Size #n
Embedded Origin Location #1
.
.
.
Embedded Origin Location #n
Embedded Dimensionality: (unsigned 32-bit integer)
This value is the number of dimensions of the space the dataset is located within. i.e. a planar dataset located within a 3-D space, or a 3-D dataset which is a subset of another 3-D space, etc.
Embedded Dimension Size: (unsigned 32-bit integer)
These values are the sizes of the dimensions of the embedded data-space that the dataset is located within.
Embedded Origin Location: (unsigned 32-bit integer)
These values comprise the location of the dataset's origin within the embedded data-space.
[Comment: need some way to handle different orientations of the dataset data-space within the embedded data-space]

HDF5 Data-Space Structured/Regular Grid Information
byte byte byte byte
Logical Dimension Size #1
Logical Dimension Maximum #1
.
.
.
Logical Dimension Size #n
Logical Dimension Maximum #n

The elements of the dimensionality message are described below:
Logical Dimension Size #n: (unsigned 32-bit integer)
This value is the current size of the dimension of the data as stored in the file. The first dimension stored in the list of dimensions is the slowest changing dimension and the last dimension stored is the fastest changing dimension.
Logical Dimension Maximum #n: (unsigned 32-bit integer)
This value is the maximum size of the dimension of the data as stored in the file. This value may be the special value <UNLIMITED> which indicates that the data may expand along this dimension indefinitely.

HDF5 Data-Space Structured/Irregular Grid Information
byte byte byte byte
# of Grid Points in Dimension #1
.
.
.
# of Grid Points in Dimension #n
Data-Type of Grid Point Locations
Location of Grid Points in Dimension #1
.
.
.
Location of Grid Points in Dimension #n

HDF5 Data-Space Unstructured Grid Information
byte byte byte byte
# of Grid Points
Data-Type of Grid Point Locations
Grid Point Locations
.
.

Examples:

Need some good examples, this is complex!

Name: Data Type

Type: 0x0003
Length: variable
Status: One required per dataset

The data type message defines the data type for each data point of a dataset. A data type can describe an atomic type like a fixed- or floating-point type or a compound type like a C struct. A data type does not, however, describe how data points are combined to produce a dataset. Data types are stored on disk as a data type message, which is a list of data type classes and their associated properties.

Data Type Message
byte byte byte byte
Type Class and Version Class Bit Field
Size in Bytes (4 bytes)


Properties


The Class Bit Field and Properties fields vary depending on the Type Class, which is the low-order four bits of the Type Class and Version field (the high-order four byte are the version which should be set to the value one). The type class is one of: 0 (fixed-point number), 1 (floating-point number), 2 (date and time), 3 (text string), 4 (bit field), 5 (opaque), 6 (compound). The Class Bit Field is zero and the size of the Properties field is zero except for the cases noted here.

Bit Field for Fixed-Point Numbers (Class 0)
Bits Meaning
0 Byte Order. If zero, byte order is little-endian; otherwise, byte order is big endian.
1, 2 Padding type. Bit 1 is the lo_pad type and bit 2 is the hi_pad type. If a datum has unused bits at either end, then the lo_pad or hi_pad bit is copied to those locations.
3 Signed. If this bit is set then the fixed-point number is in 2's complement form.
4-23 Reserved (zero).

Properties for Fixed-Point Numbers (Class 0)
Byte Byte Byte Byte
Bit Offset Bit Precision

Bit Field for Floating-Point Numbers (Class 1)
Bits Meaning
0 Byte Order. If zero, byte order is little-endian; otherwise, byte order is big endian.
1, 2, 3 Padding type. Bit 1 is the low bits pad type, bit 2 is the high bits pad type, and bit 3 is the internal bits pad type. If a datum has unused bits at either or between the sign bit, exponent, or mantissa, then the value of bit 1, 2, or 3 is copied to those locations.
4-5 Normalization. The value can be 0 if there is no normalization, 1 if the most significant bit of the mantissa is always set (except for 0.0), and 2 if the most signficant bit of the mantissa is not stored but is implied to be set. The value 3 is reserved and will not appear in this field.
6-7 Reserved (zero).
8-15 Sign. This is the bit position of the sign bit.
16-23 Reserved (zero).

Properties for Floating-Point Numbers (Class 1)
Byte Byte Byte Byte
Bit Offset Bit Precision
Exponent Location Exponent Size in Bits Mantissa Location Mantissa Size in Bits
Exponent Bias

Bit Field for Strings (Class 3)
Bits Meaning
0-3 Padding type. This four-bit value determines the type of padding to use for the string. The values are:
0 Null terminate.
A zero byte marks the end of the string and is guaranteed to be present after converting a long string to a short string. When converting a short string to a long string the value is padded with additional null characters as necessary.

1 Null pad.
Null characters are added to the end of the value during conversions from short values to long values but conversion in the opposite direction simply truncates the value.

2 Space pad.
Space characters are added to the end of the value during conversions from short values to long values but conversion in the opposite direction simply truncates the value. This is the Fortran representation of the string.

3-15 Reserved.
These values are reserved for future use.
4-7 Character Set. The character set to use for encoding the string. The only character set supported is the 8-bit ASCII (zero) so no translations have been defined yet.
8-23 Reserved (zero).

Bit Field for Compound Types (Class 6)
Bits Meaning
0-15 Number of Members. This field contains the number of members defined for the compound data type. The member definitions are listed in the Properties field of the data type message.
15-23 Reserved (zero).

The Properties field of a compound data type is a list of the member definitions of the compound data type. The member definitions appear one after another with no intervening bytes. The member types are described with a recursive data type message.

Properties for Compound Types (Class 6)
Byte Byte Byte Byte


Name (null terminated, multiple of eight bytes)


Byte Offset of Member in Compound Instance
Dimensionality reserved
Dimension Permutation
Reserved
Size of Dimension 0 (required)
Size of Dimension 1 (required)
Size of Dimension 2 (required)
Size of Dimension 3 (required)


Member Type Message


Data type examples are here.


Name: Data Storage - Fill Value

Type: 0x0004
Length: varies
Status: Optional, may not be repeated.

The fill value message stores a single data point value which is returned to the application when an uninitialized data point is read from the dataset. The fill value is interpretted with the same data type as the dataset. If no fill value message is present then a fill value of all zero is assumed.

Fill Value Message
byte byte byte byte
Size (4 bytes)

Fill Value

Field Name Description
Size (4 bytes) This is the size of the Fill Value field in bytes.
Fill Value The fill value. The bytes of the fill value are interpreted using the same data type as for the dataset.

Name: Reserved - Not Assigned Yet

Type: 0x0005
Length: N/A
Status: N/A

Name: Data Storage - Compact

Type: 0x0006
Length: varies
Status: Optional, may not be repeated.

This message indicates that the data for the data object is stored within the current HDF file by including the actual data within the header data for this message. The data is stored internally in the "normal" format, i.e. in one chunk, un-compressed, etc.

Note that one and only one of the "Data Storage" headers can be stored for each data object.

Format of Data: The message data is actually composed of dataset data, so the format will be determined by the dataset format.

Examples:

[very straightforward]

Name: Data Storage - External Data Files

Type: 0x0007
Length: varies
Status: Optional, may not be repeated.

Purpose and Description: The external object message indicates that the data for an object is stored outside the HDF5 file. The filename of the object is stored as a Universal Resource Location (URL) of the actual filename containing the data. An external file list record also contains the byte offset of the start of the data within the file and the amount of space reserved in the file for that data.

External File List Message
byte byte byte byte
Version Reserved
Allocated Slots Used Slots

Heap Address


Slot Definitions...

Field Name Description
Heap Address This is the address of a local name heap which contains the names for the external files. The name at offset zero in the heap is always the empty string.
Allocated Slots The total number of slots allocated in the message. Its value must be at least as large as the value contained in the Used Slots field.
Used Slots The number of initial slots which contain valid information. The remaining slots are zero filled.
Reserved This field is reserved for future use.
Slot Definitions The slot definitions are stored in order according to the array addresses they represent. If more slots have been allocated than what has been used then the defined slots are all at the beginning of the list.

External File List Slot
byte byte byte byte

Name Offset (<size> bytes)


File Offset (<size> bytes)


Size

Field Name Description
Name Offset (<size> bytes) The byte offset within the local name heap for the name of the file. File names are stored as a URL which has a protocol name, a host name, a port number, and a file name: protocol:port//host/file. If the protocol is omitted then "file:" is assumed. If the port number is omitted then a default port for that protocol is used. If both the protocol and the port number are omitted then the colon can also be omitted. If the double slash and host name are omitted then "localhost" is assumed. The file name is the only mandatory part, and if the leading slash is missing then it is relative to the application's current working directory (the use of relative names is not recommended).
File Offset (<size> bytes) This is the byte offset to the start of the data in the specified file. For files that contain data for a single dataset this will usually be zero.
Size This is the total number of bytes reserved in the specified file for raw data storage. For a file that contains exactly one complete dataset which is not extendable, the size will usually be the exact size of the dataset. However, by making the size larger one allows HDF5 to extend the dataset. The size can be set to a value larger than the entire file since HDF5 will read zeros past the end of the file without failing.

Name: Data Storage - Layout

Type: 0x0008
Length: varies
Status: Required for datasets, may not be repeated.

Purpose and Description: Data layout describes how the elements of a multi-dimensional array are arranged in the linear address space of the file. Two types of data layout are supported:

  1. The array can be stored in one contiguous area of the file. The layout requires that the size of the array be constant and does not permit chunking, compression, checksums, encryption, etc. The message stores the total size of the array and the offset of an element from the beginning of the storage area is computed as in C.
  2. The array domain can be regularly decomposed into chunks and each chunk is allocated separately. This layout supports arbitrary element traversals, compression, encryption, and checksums, and the chunks can be distributed across external raw data files (these features are described in other messages). The message stores the size of a chunk instead of the size of the entire array; the size of the entire array can be calculated by traversing the B-tree that stores the chunk addresses.

Data Layout Message
byte byte byte byte
Version Dimensionality Layout Class Reserved
Reserved

Address

Dimension 0 (4-bytes)
Dimension 1 (4-bytes)
...

Field Name Description
Version A version number for the layout message. This documentation describes version one.
Dimensionality An array has a fixed dimensionality. This field specifies the number of dimension size fields later in the message.
Layout Class The layout class specifies how the other fields of the layout message are to be interpreted. A value of one indicates contiguous storage while a value of two indicates chunked storage. Other values will be defined in the future.
Address For contiguous storage, this is the address of the first byte of storage. For chunked storage this is the address of the B-tree that is used to look up the addresses of the chunks.
Dimensions For contiguous storage the dimensions define the entire size of the array while for chunked storage they define the size of a single chunk.

Name: Reserved - Not Assigned Yet

Type: 0x0009
Length: N/A
Status: N/A
Purpose and Description: N/A
Format of Data: N/A

Name: Reserved - Not Assigned Yet

Type: 0x000A
Length: N/A
Status: N/A
Purpose and Description: N/A
Format of Data: N/A

Name: Data Storage - Filter Pipeline

Type: 0x000B
Length: varies
Status: Optional, may not be repeated.

Purpose and Description: This message describes the filter pipeline which should be applied to the data stream by providing filter identification numbers, flags, a name, an client data.

Filter Pipeline Message
byte byte byte byte
Version Number of Filters Reserved
Reserved

Filter List

Field Name Description
Version The version number for this message. This document describes version one.
Number of Filters The total number of filters described by this message. The maximum possible number of filters in a message is 32.
Filter List A description of each filter. A filter description appears in the next table.

Filter Pipeline Message
byte byte byte byte
Filter Identification Name Length
Flags Client Data Number of Values

Name


Client Data

Padding

Field Name Description
Filter Identification This is a unique (except in the case of testing) identifier for the filter. Values from zero through 255 are reserved for filters defined by the NCSA HDF5 library. Values 256 through 511 have been set aside for use when developing/testing new filters. The remaining values are allocated to specific filters by contacting the HDF5 Development Team.
Name Length Each filter has an optional null-terminated ASCII name and this field holds the length of the name including the null termination padded with nulls to be a multiple of eight. If the filter has no name then a value of zero is stored in this field.
Flags The flags indicate certain properties for a filter. The bit values defined so far are:
bit 1
If set then the filter is an optional filter. During output, if an optional filter fails it will be silently removed from the pipeline.
Client Data Number of Values Each filter can store a few integer values to control how the filter operates. The number of entries in the Client Data array is stored in this field.
Name If the Name Length field is non-zero then it will contain the size of this field, a multiple of eight. This field contains a null-terminated, ASCII character string to serve as a comment/name for the filter.
Client Data This is an array of four-byte integers which will be passed to the filter function. The Client Data Number of Values determines the number of elements in the array.
Padding Four bytes of zeros are added to the message at this point if the Client Data Number of Values field contains an odd number.

Name: Attribute

Type: 0x000C
Length: varies
Status: Optional, may be repeated.

Purpose and Description: The Attribute message is used to list objects in the HDF file which are used as attributes, or "meta-data" about the current object. An attribute is a small dataset; it has a name, a data type, a data space, and raw data. Since attributes are stored in the object header they must be relatively small (<64kb) and can be associated with any type of object which has an object header (groups, datasets, named types and spaces, etc.).

Attribute Message
byte byte byte byte
Version Reserved Name Size
Type Size Space Size

Name


Type


Space


Data

Field Name Description
Version Version number for the message. This document describes version 1 of attribute messages.
Name Size The length of the attribute name in bytes including the null terminator. Note that the Name field below may contain additional padding not represented by this field.
Type Size The length of the data type description in the Type field below. Note that the Type field may contain additional padding not represented by this field.
Space Size The length of the data space description in the Space field below. Note that the Space field may contain additional padding not represented by this field.
Reserved This field is reserved for later use and is set to zero.
Name The null-terminated attribute name. This field is padded with additional null characters to make it a multiple of eight bytes.
Type The data type description follows the same format as described for the data type object header message. This field is padded with additional zero bytes to make it a multiple of eight bytes.
Space The data space description follows the same format as described for the data space object header message. This field is padded with additional zero bytes to make it a multiple of eight bytes.
Data The raw data for the attribute. The size is determined from the data type and data space descriptions. This field is not padded with additional zero bytes.

Name: Object Name

Type: 0x000D
Length: varies
Status: Optional, may not be repeated.

Purpose and Description: The object name or comment is designed to be a short description of an object. An object name is a sequence of non-zero ('\0') ASCII characters with no other formatting included by the library.

Name Message
byte byte byte byte

Name

Field Name Description
Name A null terminated ASCII character string.

Name: Object Modification Date & Time

Type: 0x000E
Length: fixed
Status: Optional, may not be repeated.

Purpose and Description: The object modification date and time is a timestamp which indicates (using ISO-8601 date and time format) the last modification of an object. The time is updated when any object header message changes according to the system clock where the change was posted.

Modification Time Message
byte byte byte byte
Year
Month Day of Month
Hour Minute
Second Reserved

Field Name Description
Year The four-digit year as an ASCII string. For example, "1998". All fields of this message should be interpreted as coordinated universal time (UTC)
Month The month number as a two digit ASCII string where January is "01" and December is "12".
Day of Month The day number within the month as a two digit ASCII string. The first day of the month is "01".
Hour The hour of the day as a two digit ASCII string where midnight is "00" and 11:00pm is "23".
Minute The minute of the hour as a two digit ASCII string where the first minute of the hour is "00" and the last is "59".
Second The second of the minute as a two digit ASCII string where the first second of the minute is "00" and the last is "59".
Reserved This field is reserved and should always be zero.

Name: Shared Object Message

Type: 0x000F
Length: 4 Bytes
Status: Optional, may be repeated.

A constant message can be shared among several object headers by writing that message in the global heap and having the object headers all point to it. The pointing is accomplished with a Shared Object message which is understood directly by the object header layer of the library. It is also possible to have a message of one object header point to a message in some other object header, but care must be exercised to prevent cycles.

If a message is shared, then the message appears in the global heap and its message ID appears in the Header Message Type field of the object header. Also, the Flags field in the object header for that message will have bit two set (the H5O_FLAG_SHARED bit). The message body in the object header will be that of a Shared Object message defined here and not that of the pointed-to message.

Shared Message Message
byte byte byte byte
Version Flags Reserved
Reserved

Pointer

Field Name Description
Version The version number for the message. This document describes version one of shared messages.
Flags The Shared Message message points to a message which is shared among multiple object headers. The Flags field describes the type of sharing:
Bit 0
If this bit is clear then the actual message is the first message in some other object header; otherwise the actual message is stored in the global heap.
Bits 2-7
Reserved (always zero)
Pointer This field points to the actual message. The format of the pointer depends on the value of the Flags field. If the actual message is in the global heap then the pointer is the file address of the global heap collection that holds the message, and a four-byte index into that collection. Otherwise the pointer is a symbol table entry that points to some other object header.

Name: Object Header Continuation

Type: 0x0010
Length: fixed
Status: Optional, may be repeated.
Purpose and Description: The object header continuation is the location in the file of more header messages for the current data object. This can be used when header blocks are large, or likely to change over time.
Format of Data:

The object header continuation is formatted as follows (assuming a 4-byte length & offset are being used in the current file):

HDF5 Object Header Continuation Message Layout
byte byte byte byte
Header Continuation Offset
Header Continuation Length

The elements of the Header Continuation Message are described below:
Header Continuation Offset: (<offset> bytes)
This value is the offset in bytes from the beginning of the file where the header continuation information is located.
Header Continuation Length: (<length> bytes)
This value is the length in bytes of the header continuation information in the file.

Examples:

[straightforward]

Name: Symbol Table Message

Type: 0x0011
Length: fixed
Status: Required for symbol tables, may not be repeated.
Purpose and Description: Each symbol table has a B-tree and a name heap which are pointed to by this message.
Format of data:

The symbol table message is formatted as follows:

HDF5 Object Header Symbol Table Message Layout
byte byte byte byte
B-Tree Address
Heap Address

The elements of the Symbol Table Message are described below:
B-tree Address (<offset> bytes)
This value is the offset in bytes from the beginning of the file where the B-tree is located.
Heap Address (<offset> bytes)
This value is the offset in bytes from the beginning of the file where the symbol table name heap is located.

Disk Format: Level 2b - Shared Data Object Headers

In order to share header messages between several dataset objects, object header messages may be placed into the global small-data heap. Since these messages require additional information beyond the basic object header message information, the format of the shared message is detailed below.

HDF5 Shared Object Header Message
byte byte byte byte
Reference Count of Shared Header Message

Shared Object Header Message

The elements of the shared object header message are described below:
Reference Count of Shared Header Message: (32-bit unsigned integer)
This value is used to keep a count of the number of dataset objects which refer to this message from their dataset headers. When this count reaches zero, the shared message header may be removed from the global small-data heap.
Shared Object Header Message: (various lengths)
The data stored for the shared object header message is formatted in the same way as the private object header messages described in the object header description earlier in this document and begins with the header message Type.

Disk Format: Level 2c - Data Object Data Storage

The data information for an object is stored separately from the header information in the file and may not actually be located in the HDF5 file itself if the header indicates that the data is stored externally. The information for each record in the object is stored according to the dimensionality of the object (indicated in the dimensionality header message). Multi-dimensional data is stored in C order [same as current scheme], i.e. the "last" dimension changes fastest.

Data whose elements are composed of simple number-types are stored in native-endian IEEE format, unless they are specifically defined as being stored in a different machine format with the architecture-type information from the number-type header message. This means that each architecture will need to [potentially] byte-swap data values into the internal representation for that particular machine.

Data with a "variable" sized number-type is stored in an data heap internal to the HDF file [which should not be user-modifiable].

Data whose elements are composed of pointer number-types are stored in several different ways depending on the particular pointer type involved. Simple pointers are just stored as the dataset offset of the object being pointed to with the size of the pointer being the same number of bytes as offsets in the file. Partial-object pointers are stored as a heap-ID which points to the following information within the file-heap: an offset of the object pointed to, number-type information (same format as header message), dimensionality information (same format as header message), sub-set start and end information (i.e. a coordinate location for each), and field start and end names (i.e. a [pointer to the] string indicating the first field included and a [pointer to the] string name for the last field). Browse pointers are stored as an heap-ID (for the name in the file-heap) followed by a offset of the data object being referenced.

Data of a compound data-type is stored as a contiguous stream of the items in the structure, with each item formatted according to it's data-type.


Quincey Koziol
Robb Matzke
Last modified: Thu Oct 8 09:34:16 EDT 1998