Re: How to speed up record I/O?

FWiW the FTP for its read\write activity uses only what is available to everyone, except system state [which gains nothing in speed for the FTP]. A TRCJOB of the processing will expose what functions the FTP utility uses for its I/O. That the FTP utilizes the existing interfaces available to everyone, ensures that there is no need to test only the one versus multiple interfaces that would effect the same. Thus there would be no reason to waste the effort to try to generate calls to the QDBGET* [or other QDB I/O programs] using any interface other than those provided by the HLLs.

Regards, Chuck

Mike Amos wrote:

My name is Mike Amos and, several years ago, I was reasonably
proficient at MI programming. The past several years, with C/400
and then ILE-C, I have had no need to use anything other than the
provided APIs and MI calls supported directly via ILE-C. Today,
however, I need to speed up read/write - esp. reads - for record
IO. I know that the IBM FTP program does reads much faster than
the provided record IO interfaces because FTP reads and sends my
300MB test file twice as fast as I can read it. I have tried
every reasonable combination of _Rreadx, read() and fread() that
I can think of using both native and IFS interfaces and compiler
options.

I have a few questions:

1. Has anyone found that MI calls to the IO functions,
QSYS/QGETDBx, etc. can be safely done using in user state at
security level 50?

2. Does anyone have any experience regarding faster record IO
methods such as use of RPG, MI or any ideas?

3. Accessing the system table to make the calls is not a problem
for me, but I have not been able to find documentation for the IO
control block/file control block/other control block definitions
needed to set up the IO.

<<SNIP>>