Performance of AIX 1.2

[Jack F. Vogel]

In article <1990Aug28.012042.21056 at jdyx.UUCP> shawn at jdyx.UUCP (Shawn Hayes) writes:
>
>  I'm trying to get the performance of a PS/2 model 70 running AIX 1.2
>up and I'm looking for pointers.  Currently my system has 6 megs of
>memory and 2 megs of paging space. 
>
>The method of gauging performance is an in-house benchmark that consists of
>a PC sending transactions to the PS/2.  The PS/2 has a program that reads
>the transactions from the com port and then passes them onto a second program.
>The first program then acknowledges the txn to the PC and waits for another
>txn.  The second program reads the txn from a queue and puts the information
>into a database.
>
>The performance is being gauged by running the same benchmarks under OS/2 and
>AIX 1.2 on the same hardware.  Currently we are finding OS/2 to run about
>twice as fast as AIX 1.2.   Part of the problem is that for this test we
>want all writes to be posted to disk immediately so the test uses an open
>call with sync mode for each of the three files accessed.  
>
>I've increased the buffer size, the msg table size, and total message space
>but the numbers still don't match the OS/2 performance.
 
I don't know that I have "the answer" for you but I do have a number of
comments. First off, I would question the validity of comparing OS/2 and AIX
anyway, after all how can one compare half an OS with a whole one :-}? Sorry,
just couldn't resist.

Now, on a more serious note one of the reasons that the comparison is a problem
is that you have potentially many more processes competing for CPU time on
AIX. So, you should take a look at what else is running on the system when
you perform this test. You might want to kill any processes not essential to
your test and the system operation. Next, it sounds to me like there is a
good chance that the 2 processes on AIX are spending a lot of time sleeping,
if so there is also a chance that one or more are getting swapped out. That
would mean you are seeing time chewed up bringing them in when another serial
interrupt happens. I would take a look at a 'ps' and see if the swapper is
getting much time, or, although it will effect your test, you might run sar
a bit and see what your paging rate looks like just to get some ideas. If
your programs are getting swapped out and you have the source you could
make use of the plock() system call to lock them in core.

Anyway, those are some initial thoughts. But I would still have to ask what
the point is, so you can strip the system down to bare bones and get it to
do single-tasking file I/O as fast as OS/2, so what?? The right tool for
the right job, or the right benchmark, as they say :-} :-}!

Disclaimer: These ramblings are my own, not my employers.


-- 
Jack F. Vogel			jackv at locus.com
AIX370 Technical Support	       - or -
Locus Computing Corp.		jackv at turnkey.TCC.COM