cache OR oracle

tanyiqiang

我也很想知道。
数组的定位好像有直接计算、顺序查找、树查找、散列法等，按资料介绍CACHE用的是稀疏数组，不知道是怎么存储和查找的。

isctiger

Cache'联机帮助中查到的：
Physical Structure of Globals
Globals are stored within physical files using a highly optimized structure. The code that manages this data structure is also highly optimized for every platform that Caché runs on. These optimizations ensure that operations on globals have high throughput (number of operations per unit of time), high concurrency (total number of concurrent users), efficient use of cache memory, and require no ongoing performance-related maintenance (such as frequent rebuilding, re-indexing, or compaction).
The physical structure used to store globals is completely encapsulated; applications do not worry about physical data structure in any way.
Globals are stored on disk within a series of data blocks; the size of each block (typically 8KB) is determined when the physical database is created. To provide efficient access to data, Caché maintains a sophisticated B-tree-like structure that uses a set of pointer blocks to link together related data blocks. Caché maintains a buffer pool — an in-memory cache of frequently referenced blocks — to reduce the cost of fetching blocks from disk.
While many database technologies use B-tree-like structures for data storage, Caché is unique in many ways:
The storage mechanism is exposed via a safe, easy-to-use interface.
Subscripts and data are compressed to save disk space as well as valuable in-memory cache space.
The storage engine is optimized for transaction processing operations: inserts, updates, and deletes are all fast. Unlike relational systems, Caché never requires rebuilding indices or data in order to restore performance.
The storage engine is optimized for maximum concurrent access.
Data is automatically clustered for efficient retrieval.

isctiger

How Globals Are Stored
Within data blocks, globals are stored sequentially. Both subscripts and data are stored together. There is a special case for large node values (long strings) which are stored within separate blocks. A pointer to this separate block is stored along with the node subscript.
For example, suppose you have a global with the following contents:
^Data(1999) = 100
^Data(1999,1) = "January"
^Data(1999,2) = "February"
^Data(2000) = 300
^Data(2000,1) = "January"
^Data(2000,2) = "February"
Most likely, this data would be stored within a single data block with a contiguous structure similar to (the real representation is a series of bytes):
Data(1999):100|1:January|2:February|2000:300|1:January|2:February|...
An operation on ^Data can retrieve its entire contents with a minimum number of disk operations.
There are a number of additional techniques used to ensure that inserts, updates, and deletes are performed efficiently.

披发头陀

没看出来Caché的存储有什么特别的，觉得有些象oracle的cluster table？介绍太少，没法判断

isctiger

下一个Cache'单机版，联机帮助里都有。

isctiger

Oracle的Cluster是按某个规则将符合这个规则的记录尽可能存储在同一个block里，以此来提高某种特定查询条件的速度（很好理解，它的目标是减少IO）。所以，不论是"Clustering By Hash"还是"Clustering By Index",都有副作用:
1.大量的磁盘空间浪费
2.需要大量的管理（所以Oracle DBA很值钱）
3.不是灵活的方案，只适用于特定查询条件

而Cache'是完全不同的方案：
1.由于Cache'不存空字段，所以本身所需空间就小
2.Cache'可以通过类之间的关系选择进行集中存储,例如对父子关系,会将父对象和子对象存储在一起,对事物操作这会大大节省IO
3.Cache'对于高级用户，可以通过自定义多维数组格式来提高效率，而且对SQL和Object访问方式透明，不需修改代码

tanyiqiang

没说到CACHE内部是怎么存储和查找的。看不懂英文。
如果是我说的那些，也就明白了，如果不是那些，又是什么方法呢？

isctiger

这不就是吗：
To provide efficient access to data, Caché maintains a sophisticated B-tree-like structure that uses a set of pointer blocks to link together related data blocks. Caché maintains a buffer pool — an in-memory cache of frequently referenced blocks — to reduce the cost of fetching blocks from disk.
While many database technologies use B-tree-like structures for data storage, Caché is unique in many ways:
The storage mechanism is exposed via a safe, easy-to-use interface.
Subscripts and data are compressed to save disk space as well as valuable in-memory cache space.
The storage engine is optimized for transaction processing operations: inserts, updates, and deletes are all fast. Unlike relational systems, Caché never requires rebuilding indices or data in order to restore performance.
The storage engine is optimized for maximum concurrent access.
Data is automatically clustered for efficient retrieval.

husthxd

尝试/实战一把就有印象了.

tanyiqiang

不是要我拿金山快译来对付吧？