Re: SQL*Net Performance on W95

dpark_at_sctcorp.com wrote in message <880561998.6183_at_dejanews.com>... <snip>
> Using the same machine that dual-boots Windows 95 and Windows
>NT, he was getting numbers such as 5 ms to perform an operation on NT40
>and 46 ms to perform that same operation on W95.
>
>We put a network sniffer on and it looks as if the same number of packets
>are being sent back and forth between client and server for both NT40 and
>W95, but there's an interesting lag with W95 after the server returns the
>status of the parse operation. This lag is usually about 9ms before the
>client sends another packet to the server.

Seems to me you just proved that Microsoft's Win95 architecture sucks. I doubt that there is a problem with SQL*Net (I assume you're running the same version in both NT and 95?). The code should be the same due to the way Oracle ports it between different operating systems. There may be a slight difference in the SQL*Net for NT (maybe Unicode support? - but I doubt it), but the sequence the OCI methods are called by SQL*Net should not differ between NT and Win95.

Remember that a lot of 16bit code still lives inside the dark and smelly guts of Win95. It just may be that SQL*Net for 95 hits some 16bit code in the kernel/gdi/protocol stack that requires some thunking to 32bit.

Why not try a little experiment? Back with Oracle 7.1 I was able to get most (if not all) of the OraNT 7.1 client software to work with Win95 - all it required was some changes to DLL names, e.g. ORANT71.DLL renamed to ORA71.DLL and so on. Try and use SQL*Net for NT on 95 and see if it works (and if so) what the performance is like. It's either that, or using some low level windows debugger to hook into the SQL*Net dll's and trace assembler instructions.

Or just writing Win95 off as another Microsoft broken promise... :-)

regards,
Billy Received on Thu Nov 27 1997 - 00:00:00 CST