FIle and Block size relationships ...

[Alma J Wetzker]

Ben Duncan wrote:
> No, it is the default size of the STRUCTS to use.
> basically, it is a ISAM file manager. The read/writes
> will be done in increments in 512 bytes. RAW type usage, of which
> you speak, is generally done as it's own partition type, and
> it's own buffering.
> 
> The way to do such btree stuff is to do read / writes / seeks, but in
> your own cache and bypass the fread / fwrite / fseek stuff which uses
> the OS cache.
> 
> Alma J Wetzker wrote:
> <SNIP>
> 
>>I don't understand.  Are you asking if you should store the actual block 
>>on the disk?  That would go back to Oracle's raw file system type of 
>>stuff, a real headache, but fast.  How would you handle backup and 
>>restore?  Reloading data?
>>
>>    -- Alma
>>_______________________________________________

That makes sense then.  When dealing with memory structures, DOS (and by 
extension, Windoze) used to have a Intel quirk that would not allow more 
than one intersegment gap.  That means that any size other than a power 
of two would limit you to two 64k segments, max.  You could code around 
it by changing default segment size, but that was way icky, and caused 
lots of other stuff to break.  I haven't played with linux memory 
structures to know if the same problem exists.

     -- Alma