Re: Streams in Ada

[Ted Dennison<dennison@telepath.com>]

In article <XRrx6.7012$u93.987018@news6-win.server.ntlworld.com>, chris.danx
says...
>
>Direct IO handles single type files, i.e. all file elements are the same.  I
>want different types in the file i.e. heterogeneous files.  Streams allow
>this so i need streams.
>
>I want to allow the reading and writing of 3 or 4 types.
>
>    type Superblock_type;
>    type Inode_type;
>    type Data_block_type;
>    type FileInfo_block_type;
>
>surely then i want streams.  I did think about direct io but i'd have severe
>performance losses in the conversion to the smallest unit -- in this case
>byte_type;  Maybe i am looking in the wrong place.

Just a bit of advice. You could just do a "Superblock_Type'Read (...)" when
reading one of these objects from the stream file. However, that will cause one
Ada.Streams.Read call for each and every component in Superblock_Type. What
would be much quicker would be to create your own 'Read routine for
Superblock_Type, which makes only one call to Ada.Streams.Read for the entire
structure. Its a bit more work though, and has to be debugged, so you should
weigh that against your desire for quicker reads.

As an example of this, we used streams in a real-time simulation I'm working on
to save all the model data for a possible simulation restore later. For most of
the models we found no performance problem with just taking the default 'Read
and 'Write. However, the I/O models have significantly more data than other
models, and run at the highest rate. They were taking so much time to complete
the operation (about 20ms) that the simulation could do nothing else but save IO
data. Rewriting their 'Write and 'Read brought the time down into the trivial
level (about 100ns).

---
T.E.D.    homepage   - http://www.telepath.com/dennison/Ted/TED.html
          home email - mailto:dennison@telepath.com