GEOS SDK TechDocs
|
| 
Local Memory
| 
2 Structure of a Local Memory Heap
The GEOS memory manager is designed to deal with blocks of memory which are measured in kilobytes. Every memory block needs an entry in the global handle table; each such entry takes up 16 bytes. This memory overhead is insignificant for a 10K code block but very significant for, for example, a 100 byte database entry. Furthermore, GEOS allows a limited number of global handles. If an application were to use global blocks for small amounts of data, it could use up too many handles.
On the other hand, if an application were to store many small pieces of data in a single global block, it might have to write elaborate memory-management routines. It would have to make sure that it could resize a piece of data at will and shuffle other data to make room. This could force programmers to spend a lot of their time writing support code.
For these reasons, GEOS provides local memory routines. Applications can designate a block of memory as a local-memory heap and use LMem routines to request small amounts ( chunks ) of data. The local memory library automatically shuffles chunks in that heap to make room for new requests. Applications can change the size of a chunk at will. The cost of using the LMem routines is one added layer of indirection and a small amount of memory overhead for each chunk.
The LMem routines have another advantage. They provide a uniform way of managing small pieces of memory. This lets GEOS add functionality which applications can use in a variety of ways. For example, GEOS implements an array-management scheme based on LMem chunks; this scheme comes complete with a modified Quicksort routine. Similarly, all GEOS objects are stored in object blocks, which are a special kind of local-memory heap. This makes it easy to add or delete objects dynamically.
GEOS SDK TechDocs
|
| Local Memory
| 2 Structure of a Local Memory Heap