ab5e45a01e3d0de7b4f2.newsdawg.com   
   From: "Dariusz Piatkowski"    
      
   On Wed, 13 Jul 2011 12:28:59 UTC, "Ruediger Ihle"  wrote:   
      
   > On Wed, 13 Jul 2011 11:20:14 UTC, "Lars Erdmann"    
   wrote:   
   >    
   >    
   > > What would be the "standard" version ? The version that still    
   > > ships with eCS (even across the various fixpak updates) ?   
   >    
   > Yes.   
   >    
   >     
   > > By the way: isn't the DosTmrXXX APIs just doing DevIOCTL calls    
   > > to TIMER0.SYS    
   >    
   > Not in this universe. These APIs don't know nothing about TIMER0.   
   > AFAIK, they read directly from the hardware. And since TIMER0 re-   
   > programms exactly this hardware, bogus results are returned when   
   > TIMER0 has active clients.   
   >    
   >     
   > > It looks like it would be a good idea to include hrtx-v1.0r03.zip   
   > > contents into eCS.   
   >    
   > Nope. First of all ISTR that there were a few compatibility issues    
   > with this version. Second: The modified CLOCK01.SYS is based on an   
   > outdated IBM version, which most likely doesn't support CLOCKSCALE=x   
   > and is not guarantueed to work properly with later kernels.   
   >    
   > BTW, I think ACPI offers an additional timer that could be used    
   > for this purpose. At least Windows seems to go this route in it's   
   > implemenation of the DosTmrXXX equivalents.   
      
   I have not given up on this...maybe it's just a way for me to kill    
   time...LOL...but, it's nice to learn more details about how the OS works.   
      
   Anyways, I had attempted a number of combinations here. Also tried the    
   hrtx-v1.0r03.zip contents as well. I have to say that the '100% CPU' spike was   
   a   
   bit more managable, the spikes were fewer and lasted shorter...but this really    
   is just a 'seat of the pants' feeling...so no real statistical data to back   
   this   
   up.   
      
   Anyways...here are couple of additional pieces to shed light on the situation:   
      
   1) I am running a MSI 790X-G45 motherboard, with a 4 core AMD Phenom II 965   
   CPU,   
   currently chugging along at 4 GHz speed, quote comfortable I may add, using    
   "PSD=OS2APIC.PSD /P=4 /APIC" as the SMP 'enabler'   
      
   2) the BIOS allows me to select SMP table of: 1.1 or 1.4, I have 1.4 selected   
   => no difference between 1.1 or 1.4 on my machine, no matter what kernel    
   combination I ran either, FP5, FP6 or 20050811   
      
   3) the BIOS allows me to select APIC & IOAPIC, both are selected, even though   
   I    
   do not run ACPI   
   => Firefox performance does not differ if I run in either mode   
      
   4) tried SMP mode by using "PSD=OS2APIC.PSD /P=4 /APIC" & "PSD=OS2APIC.PSD    
   /P=4", so basically excluding APIC controlls   
   => no impact   
      
   Recently, I found out that a big stability improvement was realized by moving    
   away from the "VIRTUALADDRESSLIMIT=1536" setting. Both 1024 & 2048 appear to    
   work much better. I have chosen to keep 2048 at the moment...my typical OS    
   uptimes are in the 10 day range. Currently, as I type this, I'm on 2d 16h:56m.   
      
   I realize that the specific level of fixpacks on my machine may have a huge    
   bearing on this. I am running HPFS386 with 128MB cache as well. Not running   
   JFS    
   at the moment though.   
      
   Any other suggestions? By all means, please let me know...   
      
   Thanks!   
      
      
   --- Internet Rex 2.31   
    * Origin: NewsGuy - Unlimited Usenet $19.95 (1:261/20.999)