SuperSPARCvs HyperSPARC and SS20 stability

Hi,

If some of you remember my numerous posts, I have two MP sparcs I run 
NetBSD on since several years. I use to run different NetBSD releases on 
each, to have different testing platforms, so it is, up to know, "intended".

My starting situation, past year was:

SS10 with 2x SuperSPARCII 75MHZ and NetBSD 8.x : rock stable
SS20 with 2x HyperSPARC 125MHz and NetBSD 9.3: stable, but would, after 
a couple of days of use or compilation, start to have memory issues 
which required a reboot (system would respond to an open shell, but 
would not spam new processes)
Even before that I had a slower Ross module with 2 CPUs on a module 
(originally wanted to make a 4x system) but the module fried becoming 
unstable. Never found in years on ebay a similar module, so ended up 
buying two matched modules instead.

Then I upgraded to NetBSD 9.4 and 10.0 respectively (each system +1) and 
the SS10 continued to run rock stable, while the SS20 became unusable, 
up to freezing during boot or shortly thereafter. Refer to several posts 
here.

To be precise:
[     1.000000] cpu0 at mainbus0: mid 8: Ross,RT625 @ 125 MHz, on-chip FPU
[     1.000000] cpu0: 256K byte write-back, 64 bytes/line, sw flush: 
cache enabled
[     1.000000] cpu1 at mainbus0: mid 10: Ross,RT625 @ 125 MHz, on-chip FPU
[     1.000000] cpu1: 256K byte write-back, 64 bytes/line, sw flush: 
cache enabled

[     1.000000] cpu0 at mainbus0: mid 8: TI,TMS390Z55 @ 75 MHz, on-chip FPU
[     1.000000] cpu0: physical 20K instruction (64 b/l), 16K data (32 
b/l), 1024K external (32 b/l): cache enabled
[     1.000000] cpu1 at mainbus0: mid 10: TI,TMS390Z55 @ 75 MHz, on-chip FPU
[     1.000000] cpu1: physical 20K instruction (64 b/l), 16K data (32 
b/l), 1024K external (32 b/l): cache enabled

If I can get ROM information from dmesg or prompt, I'll give that, 
otherwise I need to hook up serial to get it.

Hard Disks of SS10 and SS20 are incompatible. But what I could do is 
swap CPU modules.

the SS10 with the HyperSparcs seems stable (stressed it a bit less, 
since I finished most pkgsrc and was compiling other things)
the SS20 with the SuperSparcs is reasonably stable, but processes will 
hang. A bit of killing and compilation can resume.


Martin and others described the process hanging as a long time issue and 
dependenton CPU speed... however, I do not (and did not) experience it 
on the SS10 with 8x and 9.x while building about the same pkgsrc 
packages. Why instead on the SS20? is it due to NetBSD 10 making things 
worse or are there other issues?

I would argue that the SUN and ROSS modules are about equivalent in real 
speed. The ROSS has higher clock, less cache... while compiling they 
feel the same. So either it is the Clock speed / Bus speed or there are 
other factors?

The SS10 and SS20 are not identical, different ROM versions and 
different BUS speeds... and of course different NetBSD OS versions.

So.. is NetBSD 10 making things worse? or is the SS20 having specific 
issues making things worse?

Riccardo