Subject: Re: Performance
To: Robert Black <r.black@ic.ac.uk>
From: Peter Berg <root@Kyra.FutureGroove.de>
List: port-arm32
Date: 04/11/1997 12:35:50
> On Apr 9, 12:00pm, Boris Boesler wrote:
> > Subject: Performance
> > Hi!
> >
> >  I upgraded from an ARM710/50MHz to a StrongARM110/202MHz. To see
> > the speed improvement I took the compile time for a small compiler
> > project. The times for generating the code generator und compiling
> > 22 .c files are listed below. I have 20+2MB RAM, 1GB SCSI-][ with
> > a Connect32 card.
> >
> > Proc	X11		terminal
> > ----------------------------------
> > SA110	3min 44sec	3min 42sec	(kernel #4958)
> > 710	4min 28sec	3min 53sec	(kernel #4740)
> >
> >  I made only one run, because the results shocked me.
> >
> >  The question: When will there be a gcc with instruction scheduling?
> 
> How will that help?
> 
> For a small compile everything is quite likely to be dominated by I/O
> performance. This is likely to be bad as I don't think we've found a way of
> turning on interrupts on the Connect32 (I could be wrong on this). Polling
> drivers are an absolute pain in OSes that do preemptive multitasking and there
> isn't much we can do about this until we figure out how to enable interrupts on
> the Connect32.

Take a look at the benchmark results. It seems not that this disappointing
behavior of a SA machine is caused by the HD controller.

                                             ARM710         SA110

File Read  (10 seconds)                   10210.0 KBps  65855.0 KBps
File Write (10 seconds)                     600.0 KBps    600.0 KBps
File Copy  (10 seconds)                     426.0 KBps    461.0 KBps 
File Read  (30 seconds)                   12351.0 KBps  64031.0 KBps 
File Write (30 seconds)                     400.0 KBps    466.0 KBps
File Copy  (30 seconds)                     520.0 KBps    676.0 KBps
C Compiler Test                              16.2 lpm       13.9 lpm

As you can see, the disc performance is the same on both machines,
nevertheless the SA equipped has less compile performance then the 'old'
ARM710.

So the HD access can not be the reason for this.



Ciao, Peter

/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
|                                                                          |
|                            >>> FutureGroove <<<                          |
|                                                                          |
| Peter Berg                                       Music production &      |
|                                                  softwaredevelopment     |
| w3: http://www.ww.tu-freiberg.de/~pberg                                  |
| email: Peter.Berg@ww.tu-freiberg.de                                      |
|                                           part time system administrator |
|                                                   at *.ww.tu-freiberg.de |
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/